MCP App Store
by KeenEthics
SitesAI Site Management icon

SitesAI Site Management MCP Server for ChatGPT

By SitesAI

Overview

SitesAI Site Management helps construction teams review company activity, find jobs and records, manage leads, CRM contacts and tasks, plan crew shift offers, review recorded shifts and signed dockets, manage workflow definitions and invitation links, and inspect site document compliance. It also supports draft quotes, contract documents, scope of work and progress-claim reports. Access is limited to the authenticated SitesAI company and requires a SitesAI account with the AI add-on. Changes require write authorization; consequential changes and notifications require confirmation. Signing dockets and sending quotes are completed in SiteAI. General attachment bodies and maintenance-history imports are not available through this public plugin. Field-capture photos and existing voice transcripts can be read without modifying the capture. SitesAI does not process payments or provide an accounting ledger.

How to connect

  1. ChatGPT: Open the SitesAI Site Management listing in the ChatGPT app directory and choose Connect. ChatGPT listing SitesAI Site Management exposes 99 tools here.

Tools

add_comment

ChatGPT
Add an activity note to any entity (requires write scope) — leads (CrLeads), tasks (TaTasks), sites (CoSites), accounts (CrAccounts), contacts (CrContacts), items, POs, members. Use to log call outcomes: the note appears in the app like a human comment. follow_up_at resurfaces tasks/leads at that time.

add_member

ChatGPT
Add a person to the company (requires write scope): creates their login record and a PENDING connection — they get an email (and a phone notification) asking them to accept, and the invitation waits for them on their SiteAI home until they accept or decline (member_added). They are not on the team, and nothing is shared with them, until they accept. engagement_type Employee (default) or Contractor (their own business — employer_company). A pending member can already be assigned a flow (assign_flow) and put on a site. Licence cap errors are returned as-is. For a SUBCONTRACTOR COMPANY: add its owner as a Contractor; once they accept, grant_site_company puts their company on your site.

add_quote_line

ChatGPT
Add a line to a DRAFT quote (requires write scope). Reference a service_id/item_id for catalog description + rate, or free text with an explicit unit_rate (ex GST).

assign_to_site

ChatGPT
Assign members and/or equipment to a site the company owns or is granted on (requires write scope). member_emails adds site memberships with site_role (default operator); already-active memberships are reported without change. unit_ids relocates equipment to this site, replacing its previous site or member-custody assignment. The site host can see assigned members, their prestarts and shared documents. Confirm equipment relocation before calling.

attach_contact

ChatGPT
Attach a contact to a SITE or a TASK (requires write scope; leads use attach_lead_contact). role: client | site_contact | receiver (the person at a delivery stop) | approver. share_live = they may watch the vehicle on this record (customer portal / tracking link); notify = departure/arrival/delivery mail. Attaching again updates role and flags. The contact must belong to this company (list_contacts / create_contact).

attach_lead_contact

ChatGPT
Attach an existing CRM contact to a lead (requires write scope). A lead can carry many contacts from different companies; optionally flag one as the primary.

calendar_events

ChatGPT
The company calendar in one call: planned shifts, task due dates, lead follow-ups and document expiries for a date window (plus what the /calendar page shows). Use for 'what's on next week?'.

call_brief

ChatGPT
THE ONE CALL TO MAKE WHEN A PHONE RINGS (g240): resolves the number once (last 9 digits, like lookup_by_phone) and returns brief — PRE-RENDERED PROSE (≤ ~1800 chars) to read, not JSON to parse: who is calling (contact + company, or one of our own team), their jobs with stage / value / quote status, follow-up dates, what we are waiting on, and the last 3 diary lines across the lead AND the contact record (inbound mail lands on the contact when it could not be matched to a job). Also returns the ids (contact_ids, lead_ids, primary_lead_id) for add_comment / create_lead afterwards. One bounded query set — safe while the caller is on the line. An unknown number returns known:false and a one-line brief, never an error. For a rare deep question mid-call use get_job_detail(lead_id).

cancel_shift

ChatGPT
Cancel an offered/accepted shift (requires write scope) — the affected member is notified (email + app push) that the shift is off.

company_overview

ChatGPT
One-call company snapshot: dashboard stats, lead pipeline (open/due/won + counts by stage), task buckets, the company timezone (all date/time strings across these tools are wall-clock in this zone; a site with its own timezone overrides it for worktime/shifts — see list_sites) and its locale (country, currency, tax_label, week_start, units, paper, calendar — EVERY money amount across these tools is in locale.currency). Use for "what is going on" briefings.

compliance_overview

ChatGPT
The company's compliance at a glance: expiring documents (own + shared with it, active holders only), submissions awaiting your review (as a site owner), documents still missing for you (as a subcontractor), and the granted sites where you have gaps.

create_account

ChatGPT
Create a CRM account (a client/supplier company) (requires write scope). Returns the new account_id.

create_contact

ChatGPT
Create a CRM contact (a person) (requires write scope). Optionally under an account. Returns the new contact_id.

create_contract

ChatGPT
Issue the CONTRACT DOCUMENT for a job (requires write scope): builds one document from the accepted quote, the approved scope of work, the variations and the claims so far, and FREEZES it. Refused when the job has neither an accepted quote nor any approved scope. Preview it first with list_contracts preview:true. After issuing, send it to the client to sign with send_contract_for_signature.

create_document

ChatGPT
Record a document (requires write scope) — for the company (omit holder/unit), one of its members (holder_email) or one of its machines (unit_id): type, title, number, issued/expiry dates. No file here (the file is attached in the app); a document with an expiry date is enough for the compliance report and the add_document / site_documents steps. Then submit_document to a site requirement.

create_job_site

ChatGPT
Give a job a site when it has none (requires write scope). A won job normally gets its site automatically; this is the manual path for a job won before that existed, for a job whose automatic creation was skipped because the company was at its LICENCE CAP, or for a second location. The site is created from the job's own name and address and linked back to it. To add further sites with their own details use the site tools. HOW A JOB IS PUT TOGETHER: a LEAD is the job — while it is being chased it is a sales lead, and the moment it moves to a WON stage it is an active job (list_jobs). The job keeps everything commercial: the accepted quote, the SCOPE OF WORK and its variations (list_sow_items), progress claims (list_progress_claims) and the contract document (list_contracts). A SITE is WHERE the job happens, linked back by lead_id: sites carry the physical things — geofenced clock-in, NFC tags, inductions, site document requirements and subcontractor grants. A job gets its site automatically when it is won, and can have several when the work spans locations. So: to staff a job or ask who is on it, work through the job (get_job); the site is where that write actually lands.

create_lead

ChatGPT
Create a new lead (requires write scope). Use when a new enquiry/prospect appears (e.g. captured on a phone call). The lead shows up in the /leads pipeline immediately.

create_quote

ChatGPT
Create a DRAFT quote on a lead (requires write scope), optionally with initial lines. Lines can reference a service_id (from list_services — fills the catalog rate) or be free text with an explicit unit_rate (ex GST). Sending stays a HUMAN action in the app, and it is a real send: pressing Send freezes the revision and EMAILS the chosen contacts the share link and a branded PDF. Where the lead has several contacts the human picks WHICH of them get it at Send time, so do not assert who will receive it. Get the lines and the contact details right before you hand it over — say the quote is ready to send, do not say it has been sent.

create_shifts

ChatGPT
Plan shifts (requires write scope): member_emails × dates on one site, each an OFFER the member accepts or rejects (email + app prompt). One member and one date = a single shift. Overlaps with existing shifts are SKIPPED with a reason, not fatal. action_ids = the DAY PLAN applied to every shift created (the operations lead their My-shift screen that day). Each member gets one summary email.

create_sow_item

ChatGPT
Add a line to a contract's scope of work (requires write scope). kind="variation" is the one to use for scope raised AFTER the contract — it starts as a DRAFT to be priced, then you move it to "submitted" (with the client) and record their answer with update_sow_item. kind="scope" (the default) is agreed works and starts approved — for the original scope prefer import_quote_sow, which copies the accepted quote's lines with their quantities and rates. Name the OPERATION that does the line (action_id) and recorded work lands on it by itself; sow_item_id on an assignment (assign_action) still works for a per-person override.

create_task

ChatGPT
Create a task (requires write scope). Defaults to assigning the company owner. Optionally link it to an entity (CoSites/CrLeads/InItems/...) so it shows on that page. Assigning to someone other than the owner notifies that person by email + app push (subject to their notification settings). DELIVERY STOPS (DDL 148): give a task an address (and optionally geo "lat, lng"; the address is geocoded when geo is absent) and it becomes a STOP — a driver whose GPS operation has gps_destinations "tasks" sees it on My shift, the app records the arrival by itself (arrived_at, an auto_bg entry with task_id), and Delivered = marking it done. Customer addresses must be tasks, never sites.

create_workflow

ChatGPT
Make a WORKFLOW (requires write scope) — ONE object for "the steps + who + where + the link": the path a worker follows on their phone, in order, from a link you send them. A workflow IS a process (flows engine) with three settings: kind (daily = persists, "Start again" each day | onboarding / once = switches off for the person when they finish), lock (locked = the person sees ONLY the workflow, their Documents and Settings until it is done or you free them | free = today's slim menu), app_required (the link shows App Store / Google Play in a plain browser — for trips / offline clock-in). EITHER give explicit steps OR let the shorthands build them in this order: documents (doc_type_ids) → induction → select_unit (+ prestart) → agree_tracking (track_consent) → clock_in (site_id) → one record_entry per action_id → clock_out. Recipes preset the shorthands: crew_onboarding = Subcontractor crew: Documents → pick the machine + prestart → agree to trips → clock in → record → clock out. Locked, needs the app. · worker_onboarding = Our worker: Documents → induction → clock in → record → clock out. Free to use the rest of the app. · operator_day = Our operator: Pick the machine + prestart → clock in → record → clock out. Free. · visitor_induction = Visitor induction: Documents → induction → clock in → clock out. Locked, once.. Then assign people/role/site and/or link to mint the join link in the same call. Returns workflow_id + links. The agent SHOULD read get_flow_design_guide once; steps are ordinary library steps (list_step_types) the office can reorder later.

delete_comment

ChatGPT
Delete a comment / activity note (requires write scope).

delete_contact

ChatGPT
Delete a CRM contact (requires write scope). Refused unless NOTHING points at it — not attached to any lead and no comments. Detach/clear those first.

delete_lead

ChatGPT
Delete a lead (requires write scope). Refused unless the lead has NO quote, linked task, comment or attached contact — clear those first.

delete_quote_line

ChatGPT
Delete a line from a DRAFT quote (requires write scope). Only works while the quote is a draft.

delete_task

ChatGPT
Delete an OPEN task (requires write scope). Refused if it has sub-tasks, comments or recorded action entries.

detach_contact

ChatGPT
Detach a contact from a SITE or a TASK (requires write scope); their tracking links for that record are withdrawn.

detach_lead_contact

ChatGPT
Detach a contact from a lead (requires write scope) — removes the link only; the contact stays in the directory. Needed before delete_lead (a lead with attached contacts refuses deletion) or delete_contact (a contact attached to leads refuses deletion).

file_capture

ChatGPT
File a NEW capture onto the record it belongs to (requires write scope). target_type: site (where the work is — the usual answer for a progress or safety photo), lead / job (the same thing: the commercial job — use it for variations, client conversations, quotes), contact (a person: a business card, a note about someone), task (evidence for one task). Filing writes a normal diary comment on that record — the worker's own words, photo, voice note and GPS, in the WORKER'S name, dated when it was captured — so the team sees it exactly as if the worker had opened the record themselves; the capture stays in the feed with your annotations. Always pass summary (one plain line saying what this shows — it becomes the comment text when the worker wrote nothing) and category; set needs_attention=true for anything unsafe, damaged, disputed or urgent. Decide from evidence: shift_site, then a nearby site within_limit, then what the photo and words say. If you are NOT sure, do not file — call update_capture with needs_attention=true and leave it new for a person. Already filed? This refuses; move it with update_capture.

get_account

ChatGPT
One CRM account in full, with its contacts and its leads.

get_capture

ChatGPT
Retrieve one company capture, its existing transcript, filing annotations, nearest-site and assigned-site hints, and up to six photos as viewable images downscaled to 1600 px. Does not generate or save transcripts, mark the capture as read, file it, or change any records. If a voice note has no existing transcript, returns transcript=null and a note directing the user to listen in SiteAI. Check the photo evidence before proposing where to file it; ai_read annotations are suggestions, not verified facts. include_image=false returns only the record and existing text.

get_docket

ChatGPT
One docket in full: header, signature + confirmation blocks, and its LINES — live entries while open, the FROZEN snapshot once signed (never re-derived). signature_path opens with get_file.

get_item

ChatGPT
Get one item by item_id — full detail incl. stock_type, for_hire / for_sale, price, unit, category, item_code and account overrides, and for tracked stock qty_on_hand (company total), avg_cost (weighted average) and stock_value. get_stock_levels for per-location quantities; list_stock_moves for the movements behind them.

get_job

ChatGPT
ONE JOB in operational detail: the site(s) the work happens at (from CoSites.lead_id), and for each one the CREW (CoSiteMember — who is assigned, their site role, whether they are inducted) and the SUBCONTRACTOR COMPANIES granted there (CoSiteCompany). This is the "who is on this job" answer. The commercial side of the same job is get_lead + list_sow_items + list_progress_claims + list_contracts. HOW A JOB IS PUT TOGETHER: a LEAD is the job — while it is being chased it is a sales lead, and the moment it moves to a WON stage it is an active job (list_jobs). The job keeps everything commercial: the accepted quote, the SCOPE OF WORK and its variations (list_sow_items), progress claims (list_progress_claims) and the contract document (list_contracts). A SITE is WHERE the job happens, linked back by lead_id: sites carry the physical things — geofenced clock-in, NFC tags, inductions, site document requirements and subcontractor grants. A job gets its site automatically when it is won, and can have several when the work spans locations. So: to staff a job or ask who is on it, work through the job (get_job); the site is where that write actually lands.

get_job_detail

ChatGPT
The NARROW deep read for one job while a caller is on the line (g240 part 2): the lead in full (follow_up_txt = the next follow-up), its contract picture (base + accepted variations), the scope-of-work item count, its quotes with status + subtotal, and the LAST 5 diary lines newest first. Four bounded queries. Use call_brief first; reach for this only when the caller asks something the brief does not cover. Same contract as REST /api/ai/v1/job-detail?leadId=.

get_lead

ChatGPT
One lead in full: every field (incl. lead_ref, the user-facing job #) plus its attached contacts (people from any company, one flagged primary), activity comments and linked tasks. THE LEAD IS THE JOB: once it reaches a WON stage it is an active job, and sites lists where the work happens (added automatically when it is won). The rest of the job hangs off this same lead_id — get_job (crew + subcontractors per site), list_sow_items (scope + variations), list_progress_claims, list_contracts.

get_maintenance_log

ChatGPT
One maintenance / prestart log in full: the header (machine, who, when, meter, fuel, site, notes, the log-level photo_path + signature_path) and its per-check LINES — result ok | defect | na | done, the note, the defect photo (photo_path → get_file) and the triage task it raised. This is the only place the per-check notes and photos are readable.

get_purchase_order

ChatGPT
Get one purchase order by po_id — supplier, deliver-to site, status (draft | quote | ordered | received | cancelled) and every line. Each line carries qty ordered and qty_received, so this answers "has the crusher dust arrived yet" and "what is still outstanding". is_stocked=false on a line means it is a service or hire charge that never enters stock.

get_quote

ChatGPT
One quote in full: header, every line (qty/unit/rate/GST, 3DX provenance), totals (ex GST / GST / inc GST), the client share URL once sent, and sent_to — the addresses this revision was actually emailed to, which on a lead with several contacts is not simply the lead contact.

get_shift

ChatGPT
One planned shift in full (status, window, note, reject reason, who responded, planned_actions = the day plan).

get_site

ChatGPT
One site in full. As the OWNER: detail (address, timezone, geofence, hours), the induction, document_requirements, granted_companies (subcontractors), members (crew of every company), units (plant on site incl. today's prestart) and offered_flows. Also lead_id/lead_name — the WON JOB (CrLeads) this site was created from, when it came from one (null for yards, workshops and offices); that is the join path from site work back to the contract. As a GRANTED subcontractor company: the site basics, the requirements that apply to you, your members/units on it and your compliance cells. Use it to check what a site_documents step will ask for, or why it skipped. Owner detail also carries the SHIFT MOVEMENTS switches (DDL 147) — track_mode (clockin = the movements of anyone clocking in here are recorded), counts_loads (an arrival here during a tracked shift = one load/delivery) — and movements: the READINESS check for trips and loads (pinned, recording switch, load-counting sites, the haul/GPS operation and who it is offered to, how many people on the site agreed to recording) with problems in plain words, in the order to fix them. Fix them with update_site (pin, switches, hours) and offer_action (the haul operation to a subcontractor company).

get_workflow

ChatGPT
ONE workflow: its settings, the ordered steps with their configs, who it is assigned to (people / roles / site offers, and which link put them there), its LINKS (audience, site, company, person, app_required, uses, active — the URL itself is shown only at mint), its RUNS of the last 30 days (open first: who is on which step, how long they have been idle there — "stuck 2 h on Documents" is the thing to act on), and the subcontractors' copies of it. Use update_workflow skip_step to move a stuck person past a step.

get_worktime

ChatGPT
ONE shift in full — THE DAY OBJECT: { shift (the list_worktime row incl. the two signatures, prestart_count, material_count), operations (the entries recorded on it — list_action_entries rows; file_path = a PDF/image attached to the operation, open with get_file), totals (per operation type), prestart (the prestart checks done on it: machine, time, defects, who), materials (every use recorded on it, oldest first; reversed = undone; reversed_after_sign = undone after the worker signed — the docket keeps it), material_totals (live quantities per item), movements (ONLY when the shift recorded its movements — DDL 147: { state: { km, moving_min, stopped_min (off site), onsite_min, visits, loads, loads_unlogged, tracking, started_at, stopped_at, stop_reason }, visits: [ { site_label, from_label, arrived_hhmm, left_hhmm, dur_min, is_load, entry_id, undone } ] } — the GPS-detected site visits; a load = an arrival at a load-counting site, written as an auto_bg operation entry; loads_unlogged > 0 = arrivals that could not be written because the shift had no SINGLE haul operation — say so) }. Pass include_track for series, the decimated GPS route as [lat, lng, at] tuples (≤ 600) — never fetch it to answer hours or loads. This is what the worker signed and the supervisor counter-signed: hours, operations, prestart, materials and movements together. A shift on YOUR site worked by a subcontractor is readable too (the plant-hire docket). Prefer this over stitching list_action_entries + list_stock_moves by work_id.

import_quote_sow

ChatGPT
Turn a quote into the contract's scope of work (requires write scope): one scope line per quote line, copying description, quantity, unit and rate. Idempotent — re-importing adds only lines that are not there yet, so it is safe to run again after the quote grows. Use it on the ACCEPTED quote once a job is won; raise anything after that as a variation (create_sow_item kind="variation").

issue_progress_claim

ChatGPT
Issue the next PROGRESS CLAIM REPORT on a contract (requires write scope). It totals the work recorded against approved scope and variations, subtracts what earlier claims reported, and FREEZES those lines so later scope edits cannot rewrite the claim. Refused when there is nothing new to claim. ⚠ This produces a REPORT (PDF at /api/claims/<claim_id>/pdf) — it does not invoice anyone, take payment or post to any ledger; SiteAI holds none. Preview first with list_progress_claims draft:true.

list_accounts

ChatGPT
List CRM accounts (client/supplier companies). The q search matches name, type, industry, site, website and phone (multi-word, order-independent).

list_action_entries

ChatGPT
Recorded KPI/action entries — the company's operation log (truck loads, pit sets, deliveries…). One row per record: entry time, qty, material, machine (text1), from_site → site (haul: load point → tip point), device geo, source (manual = driver tap · auto_fg = the haul screen's foreground auto-log at the tip · auto_bg = recorded automatically from GPS movement tracking, DDL 147: an arrival at a load-counting site, qty 1, from_site → site) and location-integrity flags (tip_far / no_location). Rows on a docket action also carry docket_id / docket_no / docket_status (DDL 122 — see list_dockets). Rows recorded on a SHIFT carry work_id (DDL 135 — the shift docket, see list_worktime); pass work_id to list one shift's operations. Use for 'how many loads were carted today / by whom / on which route'. Keyset-paged via after = last entry_id.

list_activity

ChatGPT
Company log book: EVERYTHING that happened in a time window, newest first — tasks created/done/approved, leads created & stage moves, comments (the field diary, incl. photo/file attachments), quotes created/sent/accepted, worktime clock-ins/outs, guided-flow runs, recorded KPI/action entries (truck loads etc. — details via list_action_entries), documents, maintenance logs, items, purchase orders, accounts/contacts, NFC/QR tag taps (who tapped which tag and what it did), quick captures still waiting to be filed (kind capture — a filed capture is a comment on its record). Use for "what happened in the company in the last 24h". Any event's photo_path/file_path can be opened with the get_file tool.

list_captures

ChatGPT
List field captures newest first: photos, voice notes or text dropped by workers with GPS, waiting to be filed onto a record. Default status=new; status=filed with target_type and target_id lists a record's capture history. Returns worker, capture time, note, ordered photo_paths (up to six), voice_path, existing transcript when available, location and accuracy, shift_site and nearest_site hints, filing state, summary, category, labels, needs_attention and ai_read. ai_read means SitesAI's first-pass reader supplied annotations; verify them against the evidence before filing. This tool does not generate transcripts or change captures. Pending phone uploads are omitted unless include_pending=true. after is the previous page's last capture_id.

list_contacts

ChatGPT
List CRM contacts (people). The q search matches name, role/title, email, company and phone (multi-word, order-independent). For a pure phone-number lookup, lookup_by_phone (digit matching) is still best.

list_contracts

ChatGPT
The CONTRACT DOCUMENTS issued on a job (DDL 141). In law the ACCEPTED QUOTE is already the contract — the client typed their name on the quote share page. A contract here is the consolidated PAPER: parties, the agreed price and terms from that quote, the scope of work as a schedule, the variations register and the claims to date, all FROZEN at issue so the document cannot change after it is signed. status: draft → sent (a signing link is out) → signed | declined, plus void. The signature (drawn) and the typed name land on the contract itself. PDF at /api/contracts/<contract_id>/pdf. Add preview:true to see what a NEW contract would say without issuing one. ⚠ A contract states a price because a contract must — SiteAI still does not invoice, take payments or keep a ledger.

list_dockets

ChatGPT
Signed DOCKETS (DDL 122; ONE PER TRUCK since DDL 143 — machine_key = the unit_id, else the rego the operator typed normalised (XV96OD), else '' when no machine is known; unit_label shows it): one per operator + machine + site + action + company-local day, over the KPI entries recorded that day — the plant-hire / cartage billing record. Opens automatically with the first entry on an action that has a Signature label; the customer's rep or a host supervisor SIGNS it on site (freezes the lines), the office CONFIRMS it (PDF emailed to customer + subcontractor + operator). status: open | signed | confirmed | void. You see dockets as the HOST and, if this company is a subcontractor, its own crew's dockets on other companies' sites (owner_email = the host). Signing/confirming are human actions in the app — no write tool. Keyset-paged via after = last docket_id.

list_documents

ChatGPT
The documents the company can see, in one list: its own (relation company), its members' (member — any live connection), its machines' (unit) and documents shared with it through site submissions (shared). Each row: type, holder, unit, title, doc_number, issued_on, expires_on, expiry_status (valid | expiring | expired), has_file + file_path (open with get_file). Filters: holder_email, unit_id, doc_type_id, expiry, expiring_before (YYYY-MM-DD), q. Max 500 rows.

list_files

ChatGPT
Every file attached to the company's records, newest first — comment photos / files / voice notes, field captures still waiting to be filed (source capture — once filed they are comment files on the record), operation-entry photos + PDFs + operator signatures, maintenance-log and prestart-check photos (defect evidence), shift clock-in/out photos and signatures, docket and contract signatures, documents and licence scans, inductions, item / member / contact / service / SOP pictures. Filter by record_id = ANY id the file is reachable by (a task, lead, site, shift work_id, docket, maintenance log, unit, action, entry, document, or a member's email — a shift's id returns its photos AND the photos of every load and prestart done on it), by source / kind, by who attached it (member_email) and by date. Each row carries path → open it with get_file. Use it for 'show me the photos from site X today', 'is there a signature on that docket', 'what did he say in the voice note'.

list_items

ChatGPT
List the company's items/equipment/stock. The q search matches name, part number, brand and description (multi-word). Use flag="hire" or flag="sale" to find what a customer can hire vs buy; flag="consumable" finds what gets used up on jobs (fuel, filters, aggregate); other flags: serialized, asset, kit, maintenance, low, archived. Each item exposes for_hire, for_sale, price, unit, and its STOCKING TYPE stock_type: tracked (counted stock in locations — has qty_on_hand, avg_cost (weighted average) and stock_value), non_stock (expensed when bought — plant hire, tipping, small consumables), asset (serial-numbered machine, counted by units). qty_on_hand is the company total across locations; get_stock_levels gives it per location, list_stock_moves the history and what was consumed on a job or a machine. Items may carry a barcode and units a tag_code (DDL 111) — the physical scan codes the mobile app resolves; useful to identify a specific machine a customer mentions.

list_jobs

ChatGPT
ACTIVE JOBS — the leads that have reached a WON stage, seen as work rather than as sales. Each row carries the operational picture beside the commercial one: the job's site, how many are on the crew, how many are CLOCKED IN RIGHT NOW, the contract value (agreed scope + approved variations), how much has been recorded against it, how many variations are waiting on the client, and the last claim number. Use this for "what is going on", "who is on site today", "which jobs are running". For the sales pipeline use list_leads instead. HOW A JOB IS PUT TOGETHER: a LEAD is the job — while it is being chased it is a sales lead, and the moment it moves to a WON stage it is an active job (list_jobs). The job keeps everything commercial: the accepted quote, the SCOPE OF WORK and its variations (list_sow_items), progress claims (list_progress_claims) and the contract document (list_contracts). A SITE is WHERE the job happens, linked back by lead_id: sites carry the physical things — geofenced clock-in, NFC tags, inductions, site document requirements and subcontractor grants. A job gets its site automatically when it is won, and can have several when the work spans locations. So: to staff a job or ask who is on it, work through the job (get_job); the site is where that write actually lands.

list_leads

ChatGPT
List CRM leads (sales pipeline). Filter by search text, exact stage, bucket (live/won/closed), or due=true for follow-ups due now. NOTE: a lead in the WON bucket is an ACTIVE JOB — for the work view of those (site, crew, who is clocked in, scope progress, claims) use list_jobs instead; this tool is the sales view.

list_maintenance_logs

ChatGPT
Maintenance logs per machine unit: kind = prestart (the daily checklist a worker submits before using a machine; defect lines raise a triage task) | service (a service done, meter + notes). Rows carry ok/defect/done counts, photo_count and the defect task. Filter by unit_id, kind, free text (machine / notes), date window; after = the last log_id for the next page. Open one with get_maintenance_log for the per-check lines, notes and photos.

list_members

ChatGPT
Company members that leads and tasks can be assigned to (email + display name + ui_lang = the language that person sees the app in, null = their phone decides + track_mode = whether their movements are recorded while clocked in: off | clockin — DDL 147; set on the member page in the app, no MCP writer by design, like the site clock modes). Use to resolve a person name to an email before create_task / update_task / update_lead. Customer-portal logins (clients given read-only access to their own jobs and quotes) are NOT members and never appear here — they cannot be assigned work.

list_no_shows

ChatGPT
Day-of check: ACCEPTED shifts on the site-local TODAY whose start passed the 15-minute grace with NO clock-in — 'rostered but not here'. Optionally filter to one site.

list_prestart_due

ChatGPT
Machines (asset units) that still need TODAY's prestart check: a unit with at least one active prestart definition and no prestart log yet in the company's calendar day. scope custody (default) = units ASSIGNED to the member (InItemUnits.member_email — the machine they operate; a labourer with no machine gets nothing); custody_or_site adds every unit sitting on the site that member is clocked in at. Omit member_email for the whole fleet (every due unit, whoever operates it). Prestarts are DONE on the phone — the shift screen, the machine tag or the maintenance page — and there is deliberately no MCP submit: the check needs the operator at the machine, photos, and the company's defect policy. Since DDL 142 a prestart done while clocked in lands on that shift (list_worktime rows carry prestart_count) and appears on the signed docket.

list_progress_claims

ChatGPT
PROGRESS CLAIM REPORTS for a contract (DDL 140): of the agreed scope + approved variations, how much has been done to date and what it is worth, minus what earlier claims already reported. Each row carries value_done, value_previous, value_this_claim and the frozen lines it reported. Add draft:true to see what the NEXT claim would say without issuing anything. ⚠ NOT ACCOUNTING: SiteAI issues no invoices, records no payments and keeps no ledger — a claim is a report the office sends or files, and the PDF lives at /api/claims/<claim_id>/pdf. Connecting claims to an accounting system is a future phase.

list_quotes

ChatGPT
List customer quotes (draft/sent/accepted/rejected/cancelled) with totals. Filter by status, lead, or search text. cancelled = was accepted, then fell over; rejected = they said no up front.

list_record_contacts

ChatGPT
The contacts ATTACHED to a SITE or a TASK (DDL 148) — the client, the site contact, the receiver of a delivery, an approver — with role, share_live (may watch the vehicle on it) and notify. One rule across the app: a person attached to a record sees it — a customer login whose email matches sees the record in the portal; a contact who never logs in gets a tracking link (send_tracking_link). Lead contacts are get_lead.contacts.

list_services

ChatGPT
List the company's service catalogue + pricing (for customer pricing & support). Each service: name, description, category (parent_name), and rate + rate_unit (e.g. 9.5 "/m3") and/or a price_note. Active only unless include_inactive. Operation types (the units of recorded field work) can group under a service — see list_actions.

list_shifts

ChatGPT
PLANNED shifts (the roster/calendar): offers to members with status offered | accepted | rejected (reject_reason says why) | cancelled | expired (offer passed unanswered — reassign or re-offer it). Shifts on YOUR sites planned by guest companies appear read-only (owner_email = the planning company). planned_actions lists the operations planned for that day (the day plan — plan_shift_actions). Filter by window (YYYY-MM-DD), site, member, status.

list_sites

ChatGPT
List the company's sites/work locations (projects, workshops, storage, offices). The q search matches name, address, project ref and phone (multi-word). Active only unless include_inactive. Each site may set working hours work_start_time/work_finish_time ("HH:MM"): live clock-ins before start record the start time, clock-outs after finish record the finish time (finish <= start = overnight site).

list_sow_items

ChatGPT
THE SCOPE OF WORK of a contract (the contract IS the lead — see list_leads/get_lead). Each item is either kind="scope" (the agreed works, usually imported from the accepted quote with import_quote_sow) or kind="variation" (scope added or changed later, which the client approves). Rows carry qty_planned/unit/rate, the rolled-up qty_done + entry_count + pct from operations recorded against them (TaActionEntries), and value_planned/value_done. status: draft (being priced) → submitted (with the client) → approved | rejected, then complete or void. Filter by lead_id, site_id or kind. NOTE: these values exist to VALUE PROGRESS in a report — SiteAI does not invoice, take payments or keep an accounting ledger.

list_tasks

ChatGPT
List company tasks. Filter by status (open/done/approved), assignee email, search text, or due=true for tasks due now.

list_workflows

ChatGPT
The company's WORKFLOWS (every process is one): kind, lock, app_required, step count, people on it, open runs right now, live links, the clock-in site. q = name contains; include_inactive for switched-off ones.

list_worktime

ChatGPT
Crew shifts — who worked / is clocked in where. Rows carry planned_start/finish (the accepted roster offer, for planned-vs-actual) and the signature state. THE DAY SCREEN (DDL 142): rows also carry prestart_count (prestart checks done on the shift — list_prestart_due for what is still due) and material_count (stock used on it — list_stock_moves with work_id); both sit on the signed docket with the operations. Filter by site, open/closed, date range (YYYY-MM-DD). Closed shifts carry a two-stage signature: the member signs at clock-out (is_signed, signed_txt, employee_sign = drawn-signature image path — open with get_file; a member-signed record is locked for the member), then the hours are SIGNED OFF via the QR on that page (is_sup_signed, sup_signed_txt, supervisor_sign path, optional supervisor_comment). Since 2026-08-31 ANY signed-in SiteAI user holding that QR may sign off — the named supervisor is often not on site — so the record says WHO did: supervisor_name = signer email, supervisor_display_name = the name they signed under, supervisor_company, and supervisor_relation = what they were to the worker at that moment (manager | supervisor | site = a supervisor on that site, possibly another company | member = same company | visitor = no connection to the company; null on records signed before that date). The worker can never sign off their own hours. Live times are clamped to the site's working hours and a 24h max shift. Rows may carry flags (comma list) — location-integrity markers the supervisor reviews before counter-signing: clockin_far / clockin_no_location / clockout_far / clockout_no_location (device position vs the site pin under the site's location rule), start_clamped (clock-in before the site start — the record carries the site start time, not the tap: check the site's start time if it looks wrong), finish_clamped (clock-out after site finish, time clamped), auto_closed (never clocked out — closed automatically at the site finish time), offline (recorded on the device with no connection — times are as tapped, synced later; hard-geofence and enforced-hours violations are FLAGGED not rejected for offline records: see offline_conflict and outside_hours), offline_conflict (offline event collided with an existing open shift or an unavailable site — may be a zero-length evidence row the supervisor resolves), outside_hours (offline clock-in outside the site's enforced hours), geo_auto (clocked in/out AUTOMATICALLY by the native site-radius geofence — sites with auto_clock_mode set; same referee and clamps as every other clock path). site_km / finish_site_km give the actual distances. Worktime is the SHIFT container — per-load / per-record production within a shift lives in list_action_entries. THE SHIFT IS THE DOCKET (DDL 135): operations the worker records on their shift (the My shift screen after clock-in) carry work_id = this row, entry_count says how many, and the two clock-out signatures cover them together with the hours; list_action_entries with work_id lists them. A HOST (site owner) also sees SUBCONTRACTOR shifts worked on its own sites (DDL 136) — read-only; owner_email is the subcontractor, site_owner_email the host. SHIFT MOVEMENTS (DDL 147): rows whose movements were recorded carry has_track, is_tracking (live now), track_km, track_visits, track_loads (automatic site-arrival loads), track_loads_unlogged and track_stop_reason; get_worktime → movements lists the site visits, movements_report totals them.

lookup_by_phone

ChatGPT
Who is this phone number? Matches the last 9 digits across CRM contacts, accounts, company members and leads. Call this first on every inbound call/text.

move_crew_member

ChatGPT
Move a person on one of your sites UNDER one of your subcontractors (requires write scope) — the repair for a driver who signed up as his own one-person company instead of joining his boss (get_site.granted_companies shows such companies: the owner is their only person). Does what a tap on the subcontractor's crew link does (they become its Employee, on the site under it), then the one-person company leaves you: its site grant, its site row and its Contractor connection to you are revoked (your seat frees), and the person's home becomes the subcontractor. Their past shifts stay as history. company_owner_email must be a company on the site (or an approved Contractor connection — granted here then).

movements_report

ChatGPT
MOVEMENT totals for TRACKED shifts (DDL 147) — the answer to "how many loads did Sam cart today", "how far did the trucks drive this week", "how often was the tip visited and by whom", WITHOUT listing shifts. Companies switch recording on per site (track_mode) or per member; the app then records the driver's GPS route while clocked in, detects visits to the company's pinned sites and counts an arrival at a load-counting site (counts_loads) as ONE load/delivery, written as an auto_bg operation entry (from_site → site, qty 1). Group by member (per driver), shift (one row per tracked shift, with work_id) or site (how many times each site was visited, in how many shifts, by whom, how many loads landed there, time on site). Rows and the grand total carry visits, loads, loads_unlogged, km, moving_min, stopped_min (stopped OFF site), onsite_min. loads_unlogged > 0 = arrivals counted but NOT written as operations because the shift had no single haul operation (an operation type with label_from_site) — tell the user to plan exactly one (plan_shift_actions / assign_action). Same figures as /worktime/movements in the app; per shift, get_worktime → movements.

operation_compliance

ChatGPT
WHO IS DOING AN OPERATION WITHOUT THE DOCUMENTS IT REQUIRES: every member connected to an operation type (directly or through a role — list_action_assignments) × required document type where the member's wallet has no active document of that type (missing) or only an expired one; include_expiring adds documents expiring within 30 days. member_email narrows to one person. The operation-level twin of site_compliance; read it before rostering.

review_submission

ChatGPT
Accept or reject a document submitted against a requirement on a site you own (requires write scope). submission_id comes from site_compliance rows. Rejected = the subcontractor sees the note and resubmits.

search_records

ChatGPT
Global search across ALL company records in one call — leads (job #/name/client), quotes, tasks, comments/diary notes (incl. photo/file attachments — open them with get_file), sites, members, CRM accounts & contacts, items + unit serials, purchase orders, documents, operations processes/SOPs, services. Multi-word: every word must match somewhere ('john pool' finds John's pool job). Use it when you don't know which entity a name/number belongs to. Returns grouped hits, each with an app URL.

send_contract_for_signature

ChatGPT
Email the CLIENT a one-time link (/c/<token>, 30 days) to read and SIGN a contract (requires write scope). They see the works, the schedule, the variations and the price, then sign with a drawn signature and their typed name — which lands on the contract record (status signed | declined, with who and when). Moves a draft contract to "sent". 30 days rather than a variation's 14, because a contract goes round a client's office.

send_variation_for_approval

ChatGPT
Email the CLIENT a one-time link (/v/<token>, 14 days) to approve or decline ONE variation (requires write scope). They read what changed, the quantity, the rate and the value, then type their name — that typed name IS their written approval and it lands back on the variation (status approved | rejected, with who and when). Moves a draft variation to "submitted" (with the client). Only variations are sent this way — agreed scope is not re-approved.

site_compliance

ChatGPT
WHO IS NON-COMPLIANT ON WHAT, right now: every requirement × subject cell on a site (company / member / machine) with the submitted document, its status (submitted | accepted | rejected | null = missing), expiry_status (valid | expiring ≤30 d | expired) and outstanding = missing/rejected/expired; plus a summary by company and awaiting_review. Owner sees every company; a granted subcontractor sees its own cells. only_outstanding=true returns just the gaps.

submit_document

ChatGPT
Submit a document against a site requirement (requires write scope) — the subcontractor side of compliance. requirement_id from list_site_document_requirements / site_compliance; member_email for a member-scope requirement (the document's holder), unit_id for a unit-scope one. Re-submitting the same cell replaces the previous submission (back to "submitted").

update_account

ChatGPT
Update a CRM account (requires write scope) — ONLY the fields you pass change. Set is_active=false to archive it.

update_capture

ChatGPT
Change a capture without filing it, or fix one that was filed wrong (requires write scope). Annotate: summary, category, labels, needs_attention (true = a person should look; the office inbox pins it). MOVE a filed capture: pass target_type + target_id — its diary comment moves with it (same comment, new record). status=dismissed (+ dismiss_reason: duplicate, pocket photo, test, personal…) takes it out of the queue and REMOVES its diary comment if it had one; status=new reopens a dismissed or filed capture (the diary comment is removed) so it can be filed again. The capture row and its files are never deleted by this tool.

update_comment

ChatGPT
Edit a comment's note text (requires write scope). The note is re-attributed to the company owner.

update_contact

ChatGPT
Update a CRM contact (requires write scope) — ONLY the fields you pass change. Set is_active=false to archive it.

update_lead

ChatGPT
Update a lead (requires write scope) — ONLY the fields you pass change. Move stage (built-ins new/open/won/awarded/completed/lost/not_qualified/closed, OR any custom stage code the company defined — see company_overview counts_by_stage; give lost_reason when losing), rename the lead, set the client company or source, change the lead type by name, assign a member (email from list_members; empty string unassigns — assigning notifies that member by email + app push, subject to their notification settings), set/clear the follow-up, priority (1=hot .. 3=low), value ($), quote dates, description, address, work type (the kind of job, e.g. 'Site-cut'), or the Drive folder / files URLs.

update_quote

ChatGPT
Record the client's decision on a quote (requires write scope), with an optional note of what they said. A SENT quote can move to accepted or rejected. An ACCEPTED quote can move to cancelled — use that, not rejected, when a job that was already won falls over; rejected means they turned it down in the first place. Reopen anything closed with Revise in the app. NOTIFIES: this emails the lead's assignee and the company owner, so do not use it to fix a typo. Sending a quote is still a human action. The LEAD stage stays the outcome system of record — update the lead too.

update_quote_line

ChatGPT
Edit a line on a DRAFT quote (requires write scope) — REPLACES the line with the values you pass (description + qty required). Changing unit_rate flips the line to a manual price so the 3DX sync never overwrites it. Only works while the quote is a draft.

update_site

ChatGPT
Change a site's settings (requires write scope; the site owner only). Partial: only the fields you pass change. geo = "lat, lng" (the pin — geocode the address first; "" clears it); clockin_mode off|soft|hard with clockin_limit_km; auto_clock_mode off|prompt|auto (needs a pin); work_start_time / work_finish_time "HH:MM" ("" clears — clock-ins BEFORE the start are recorded AS the start, so a wrong start time rewrites every morning clock-in: 15:44 did exactly that); enforce_site_hours; timezone (IANA); track_mode off|clockin (record the trips of anyone clocking in here); counts_loads (an arrival here counts as one load — the tip, or this site when trucks return here to load); is_active. Returns the site and its movements readiness so you can see what is still missing. The haul operation itself is offered to a subcontractor with offer_action; a driver agrees to recording once, on My shift.

update_sow_item

ChatGPT
Edit a scope/variation line or move it through the contract lifecycle (requires write scope): status draft → submitted (sent to the client) → approved | rejected, then complete when the work is done or void if withdrawn. Record what the client said in client_comment and who agreed in decided_by_name — the approval date is stamped automatically. A rejected line is KEPT (contract records are never deleted).

update_task

ChatGPT
Update a task (requires write scope): set status (done / open / approved — done on a delivery stop = delivered), reassign to a member (email from list_members), change/clear the due date, or set/clear the delivery-stop address and pin (address / geo; "" clears; an address without geo is geocoded). Reassigning notifies the new assignee by email + app push (subject to their notification settings).

update_workflow

ChatGPT
Change a WORKFLOW (requires write scope). Partial: only what you pass changes. name / description / kind / lock / app_required / is_active (false = switch off; runs and links die with it); steps = the WHOLE ordered list (pass sop_id to keep a step, a type+config to make a new one; steps left out are unlinked, not deleted); people_add / people_remove (member emails); role_id / site_id add an assignment; skip_step {member_email, sop_id} marks the step that person is standing on as done for them (owner rule: sequence gates, a supervisor can skip — e.g. a lost licence photo must not stop a truck). Changing lock or kind re-flags every active assignment.

update_worktime

ChatGPT
Edit a FINISHED worktime record's times/break/comment (requires write scope). RULES: a supervisor-signed record is LOCKED (refused). Editing a member-signed record CLEARS the member's signature — they must sign again, then the supervisor signs. Warn the person you're helping about that before editing a signed record. The new window must NOT overlap another record for the same member (a member cannot be on two shifts at once) — the edit is refused and names the clashing record.

void_contract

ChatGPT
Withdraw a contract (requires write scope). The row is KEPT — a contract is a record, never deleted — and any outstanding signing link stops working. A SIGNED contract can only be voided, never edited back: it is the record of what was agreed. Issue a fresh contract (create_contract) if the terms have changed.

worktime_summary

ChatGPT
Hour TOTALS for a date range — the answer to "how many hours did Sam work last month" or "how much time went into that site", WITHOUT listing shifts and adding them up yourself. Prefer this over list_worktime for any question about an amount of time; list_worktime is for the individual records. Group by member (payroll), site (job costing) or member_site (who worked where). Returns per-group rows plus one grand total: minutes, break_minutes, dockets (shifts counted), open_dockets (still clocked in), sup_signed (supervisor counter-signed), flagged, and the date span. IMPORTANT: minutes EXCLUDES any record over 24 hours — that is an abandoned clock-in, not a shift — and those are counted separately as abandoned. If abandoned > 0, say so: the real total is understated until someone fixes those records on the Worktime page. The same figures download as CSV from /worktime/report in the app.

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/sitesai-site-management"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/sitesai-site-management)