MCP App Store
by KeenEthics
Cigo Foresight icon

Cigo Foresight MCP Server for ChatGPT

By Cigo Tracker

Overview

Cigo Foresight connects ChatGPT to your Cigo Tracker account. Sign in with your Cigo login and ask about jobs, routes, drivers, invoices and delivery performance in plain language. ChatGPT reads Foresight's reporting tables with one SQL SELECT at a time, and the Cigo Dispatch API record by record, always limited to the providers your login belongs to and, for a merchant, only its own locations. If you allow writes when you sign in, ChatGPT can also create, update, cancel and delete Dispatch records. Changes take effect immediately in your Cigo account. You need a Cigo Tracker account with Foresight enabled.

How to connect

  1. ChatGPT: Open the Cigo Foresight listing in the ChatGPT app directory and choose Connect. ChatGPT listing Cigo Foresight exposes 32 tools here.

Tools

concluded_jobs_report

ChatGPT
Create a CSV file of the Concluded Jobs report (Order Management → Concluded Jobs → Export) of the signed-in Cigo user's providers, and a link that downloads it, for the page's filters; no Cigo record changes. The file has no action details, and the server writes the SQL and the file, of any size. Action details are temporarily unavailable in this report; we're working on it. If asked for them, say so and offer the report without them. include_actions must be false. start_date and end_date are the Date Range on the scheduled date, both included, at most 400 days apart. Empty filters restrict nothing, as on the page; locations are reference numbers or names, exactly; search matches any of up to 5 words. Answers {"ok": true, "export_id", "status", "rows_counted", "rows_written", "download"}, or {"ok": false, "error"} naming the argument to correct. Give the user the download link: it opens after their Cigo sign-in and downloads the file once it is whole, which a large report takes minutes to be. status is waiting, reading, done or failed, with error; export_status tells how far it got. left_out names days the file lacks.

dispatch_cancel_jobs

ChatGPT
Cancel jobs of the user's linked Dispatch account, 1 to 50 a call. It sends POST /jobs/id/{jobID}/cancel once for each ID of job_ids. job_ids are job IDs a read returned, each once. A cancel sets the job's status to cancelled and triggers Dispatch's cancellation workflows; only a supervisor sub-user may cancel. A cancelled job cannot be deleted afterwards: Dispatch answers 403 to its delete. To make a job go away, delete it instead of cancelling it. Each cancel is made at once with no undo: cancel only what the user asked for. One job answers as Dispatch did, with read_back. Several answer results in the order of job_ids, each with its ok, status_code and read_back; a result's place in results is its job's in job_ids, except one skipped after the call stopped, which names its entry. With stop_at_first_failure the first job not cancelled ends the call and skips the rest. The jobs not sent, after 30 seconds, when the answer is full or when the account's turn or minute refuses one, are left with why_left, to pass again as job_ids. ok is true when every cancel was accepted.

dispatch_check_availability

ChatGPT
Check what the user's linked Dispatch account can still book, by one of these. slots, with a date as YYYY-MM-DD and a geolocation as latitude,longitude: the jobs, weight and volume that day can still take there (GET /available-slots). booking_dates, with a geolocation or an address and the job's weight, volume or handle time when they matter: the days a job can be booked (GET /available-booking-dates). Dispatch answers 403 when the account does not have them enabled. fields keeps only the keys it names, offset and limit return part of each list with lists giving each list's total, and an answer over 40,000 characters is refused with what to pass instead. Keys of the answer, names only, none sure to be there, statusCode and status left out ([ ] a list's records' keys, { } an object's, * not in the contract): GET /available-slots: max_jobs, available_jobs_slots, max_weight, available_weight_slots, max_volume, available_volume_slots. GET /available-booking-dates: available_booking_dates. A name in fields is one of these.

dispatch_copy_jobs

ChatGPT
Copy jobs of the user's linked Dispatch account to another date, as Foresight's own chat copies them. It reads each job with GET /jobs/id/{jobID} and creates its copy with POST /jobs. job_ids are 1 to 50 job IDs a read returned, each once, and date is the copies' day as YYYY-MM-DD. Each job is read first, and nothing is sent while one cannot be read. A pickup or a dropoff Cigo generated is left out, under generated_stops: Cigo generates it again when the copy is routed. Each copy is built from the job as the create takes it, with the fields the contract accepts, the job's actions, out of staging and without the reference the create generates, and checked against the contract; a copy it would refuse is under not_copied with the reason and is not sent. The copies that pass are created one after another, as the account's writes are, each at once and with no undo. created has each source with its name, the new_id Dispatch gave the copy and its read_back, as dispatch_create_jobs answers one: verified when the copy reads back, mismatch with what was found, or unread when Dispatch did not answer the read in time; refused the copies Dispatch refused and unknown those it did not answer, which it may have made. The reads take up to 30 seconds and the copies 30 more, about ten to fifteen copies at the demo's latency: the copies not sent by then, or after the account's turn or minute refused one, are left with why_left: call again with call_again_with, exactly as it is, which keeps date. ok is true when every copy sent was accepted.

dispatch_create_jobs

ChatGPT
Create jobs in the user's linked Dispatch account, 1 to 50 a call, as an import makes them. It sends POST /jobs once for each job of jobs. Each entry of jobs is one job as the contract declares it, with its actions; two equal entries make two jobs. A job goes to staging, the Import Tool, for a dispatcher to review unless it sets skip_staging, and a sub-user must pass branch_id and distribution_center_id. Creating a job also creates a customer from the job's contact fields when the account has none that matches them; the customer stays when the job is deleted. A job with an email or a mobile_number may be notified to its customer by Email and SMS, even on the demo. Every job is checked against the contract before the first leaves; one it would refuse refuses the call with its number. Each is created at once with no undo: create only what the user asked for. One job answers as Dispatch did: data with the new job_id, and read_back. Several answer results in the order of jobs, each with its ok, status_code, read_back and the ids Dispatch named; a result's place in results is its job's in jobs, except one skipped after the call stopped, which names its entry. With stop_at_first_failure the first job not accepted ends the call and skips the rest. The jobs not sent, after 30 seconds, when the answer is full or when the account's turn or minute refuses one, are left with why_left, to pass again as jobs. With outcome_unknown Dispatch may have made the job: search for it before creating it again. ok is true when every job was accepted.

dispatch_create_route

ChatGPT
Create a route, an itinerary, in the user's linked Dispatch account: one a call. It sends POST /itineraries. route is the route as the contract declares it, with its day, start, vehicle and the jobs it takes as add_ids; a job on a route takes only some changes and cannot be deleted. A sub-user may not create a route. The route is made at once with no undo: create only what the user asked for. The answer is Dispatch's, with the new itinerary_id, and read_back says whether the route reads back holding its jobs.

dispatch_delete_jobs

ChatGPT
Delete jobs of the user's linked Dispatch account, 1 to 50 a call, or their actions. It sends DELETE /jobs/id/{jobID} once for each ID, DELETE /jobs/id/{jobID}/actions once for each ID with all_actions, or DELETE /jobs/id/{jobID}/actions/{actionID} with action_id. job_ids are job IDs a read returned, each once. Alone, each job is deleted; with all_actions, each job's actions are and the job stays; with action_id, that one action of the one job in job_ids. Dispatch deletes a job only while it is in staging or pending and on no route, and a sub-user only the jobs it created. A cancelled job cannot be deleted: Dispatch answers 403 to the delete of a job once it is cancelled, and the job stays as cancelled. A job on a route keeps its pickup, return and exchange actions, and a fulfilled job keeps all of its actions. A delete has no undo: delete only what the user asked for. One delete answers as Dispatch did, with read_back, verified once the record reads as gone. Several answer results in the order of job_ids, each with its ok, status_code and read_back; a result's place in results is its job's in job_ids, except one skipped after the call stopped, which names its entry. With stop_at_first_failure the first not accepted ends the call and skips the rest. The jobs not sent, after 30 seconds, when the answer is full or when the account's turn or minute refuses one, are left with why_left: call again with call_again_with, exactly as it is, which keeps all_actions. ok is true when every delete was accepted.

dispatch_delete_route

ChatGPT
Delete one route, an itinerary, of the user's linked Dispatch account, while it has not started yet. It sends DELETE /itineraries/id/{itineraryID}. A sub-user may not delete one. A delete has no undo: delete only what the user asked for. The answer is Dispatch's, and read_back is verified once the route reads as gone.

dispatch_delete_vehicle_breaks

ChatGPT
Delete break windows of one vehicle of the user's linked Dispatch account: 1 to 50 a call. It sends DELETE /vehicles/id/{vehicleID}/breaks/{breakID} once for each ID of break_ids, or with itinerary_id DELETE /itineraries/id/{itineraryID}/vehicles/id/{vehicleID}/breaks/{breakID}. Pass the vehicle_id and its break_ids, each once. With itinerary_id, each break comes off that route only and the vehicle keeps it. A sub-user may not delete a vehicle's breaks. A delete has no undo: delete only what the user asked for. One delete answers as Dispatch did. Several answer results in the order of break_ids, each with its ok and status_code; a result's place in results is its break's in break_ids, except one skipped after the call stopped, which names its entry. With stop_at_first_failure the first not accepted ends the call and skips the rest. The breaks not sent, after 30 seconds, when the answer is full or when the account's turn or minute refuses one, are left with why_left: call again with call_again_with, exactly as it is, which keeps itinerary_id. ok is true when every delete was accepted.

dispatch_get_customers

ChatGPT
Read customers of the user's linked Dispatch account: one by customer_id (GET /customers/id/{customerID}), or with search the first 20 that match a name, which may be partial, a phone or mobile number, an email, an invoice or a reference (POST /customers/search). Pass one of the two. fields keeps only the keys it names, offset and limit return part of each list with lists giving each list's total, and an answer over 40,000 characters is refused with what to pass instead. Keys of the answer, names only, none sure to be there, statusCode and status left out ([ ] a list's records' keys, { } an object's, * not in the contract): GET /customers/id/{customerID}: customer{customer_id, full_name, phone_number, mobile_number, address, coordinates, reference_id, defaults, jobs}. POST /customers/search: customers[customer_id, full_name, reference_id, phone_number, mobile_number, email]. A name in fields is one of these, or one key below it and no deeper.

dispatch_get_drivers

ChatGPT
Read the drivers, Dispatch's operators, of the user's linked Dispatch account: all of them without operator_id (GET /operators), or one with it (GET /operators/id/{operatorID}). fields keeps only the keys it names, operators.name of each operator, offset and limit return part of each list with lists giving each list's total, and an answer over 40,000 characters is refused with what to pass instead. Keys of the answer, names only, none sure to be there, statusCode and status left out ([ ] a list's records' keys, { } an object's, * not in the contract): GET /operators: operators[id, name, reference_id, username, phone_number, language, timezone, last_active, avg_rate, activated]. GET /operators/id/{operatorID}: operator{as operators}. A name in fields is one of these, or one key below it and no deeper.

dispatch_get_jobs

ChatGPT
Read 1 to 50 jobs of the user's linked Dispatch account by ID, one part of each. part is job, the job with its actions and the stops related to it on its route (GET /jobs/id/{jobID}); reports, its signatures, fulfillment, incidents and what was left incomplete (GET /jobs/id/{jobID}/reports); chat, its chat messages (GET /jobs/id/{jobID}/chat), or with message_id, which a message's create answered, that one message of one job with whether it was read (GET /jobs/id/{jobID}/chat/{messageID}); location, where its operator is while the job is in progress (GET /jobs/id/{jobID}/location), with the ETA and distance when route_data is true; or geofencing, its geofence entries and exits (GET /geofencing/job_id/{job_id}). For one ID the answer is the read's data, with lists giving the total of each list in it, such as the job's actions, which offset and limit cut. For more, results has one result for each job read, in the order given, with its job_id, its own ok and its own data and lists: one that Dispatch refused does not undo the others. left has the IDs still to read: send call_again_with, exactly as it is, after retry_after_seconds when it is there. Pass fields, such as ["job.status", "job.date", "actions.invoice_number"], so that all of them fit: it keeps only the keys it names; a name the part never answers is refused, or named in fields_not_found, with the keys it has. reports.incomplete.reason_approved is true by default: a report not submitted, with no message, holds no reason to accept. Keys of the answer, names only, none sure to be there, statusCode and status left out ([ ] a list's records' keys, { } an object's, not in the contract): job: job{}, actions[], related_on_route_stops[]. reports: reports{}. chat: chat[]. chat with message_id: message_id, job_id, message, sent_on, is_read. location: job{}. geofencing: geofence_events[], first_geofence_entry, last_geofence_exit, timezone. A name in fields is one of these, or one key below it and no deeper.

dispatch_get_locations

ChatGPT
Read the saved locations of the user's linked Dispatch account, its depots and other places: all of them without location_id (GET /locations), or one with it (GET /locations/id/{locationID}). fields keeps only the keys it names, offset and limit return part of each list with lists giving each list's total, and an answer over 40,000 characters is refused with what to pass instead. Keys of the answer, names only, none sure to be there, statusCode and status left out ([ ] a list's records' keys, { } an object's, * not in the contract): GET /locations: locations[id, type, name, reference_id, address, postal_code, coordinates, activated]. GET /locations/id/{locationID}: location{as locations}. A name in fields is one of these, or one key below it and no deeper.

dispatch_get_routes

ChatGPT
Read the routes, Dispatch's itineraries, of the user's linked Dispatch account by one of these. dates, the routes of 1 to 50 days as YYYY-MM-DD (GET /itineraries/date/{date}). itinerary_ids, 1 to 50 routes by ID (GET /itineraries/id/{itineraryID}), or with polyline true the path each one drives, as GeoJSON (GET /itineraries/id/{itineraryID}/polyline). operator_id or vehicle_id, the routes of an operator or a vehicle (GET /itineraries/operator/{operator}, GET /itineraries/vehicle/{vehicle}) from start_date to end_date, at most 30 days apart, or of the last 7 days without them. For several dates or itinerary_ids, results has one result for each day or route read, with its date or itinerary_id and its own ok, and left the days or IDs still to read: call again with call_again_with, exactly as it is, after retry_after_seconds when it is there. fields keeps only the keys it names, itineraries.stops of each route among them, offset and limit return part of each list with lists giving each list's total, in the answer for one day or route and in each result for several, and an answer over 40,000 characters is refused with what to pass instead. Keys of the answer, names only, none sure to be there, statusCode and status left out ([ ] a list's records' keys, { } an object's, * not in the contract): GET /itineraries/date/{date}: date, itineraries[itinerary_id, numeric_id, date, planned_departure_datetime, time, time_frame_size, start_location, end_location, route_status_code, route_error_message, round_trip, route_metrics, route_options, vehicle, operators, stops_count, stops, options]. GET /itineraries/id/{itineraryID}: itinerary{as itineraries}. GET /itineraries/id/{itineraryID}/polyline: geojson{type, coordinates}. GET /itineraries/operator/{operator} and GET /itineraries/vehicle/{vehicle}: date{start_date, end_date}, itineraries[as above]. A name in fields is one of these, or one key below it and no deeper.

dispatch_get_vehicles

ChatGPT
Read the vehicles of the user's linked Dispatch account: all of them without vehicle_id (GET /vehicles), one with it (GET /vehicles/id/{vehicleID}), or with breaks true the breaks configured on that one (GET /vehicles/id/{vehicleID}/breaks). fields keeps only the keys it names, vehicles.name of each vehicle, offset and limit return part of each list with lists giving each list's total, and an answer over 40,000 characters is refused with what to pass instead. Keys of the answer, names only, none sure to be there, statusCode and status left out ([ ] a list's records' keys, { } an object's, * not in the contract): GET /vehicles: vehicles[id, name, reference_id, type, weight_capacity, volume_capacity, license_plate, start_time, start_location, end_location, activated, breaks]. GET /vehicles/id/{vehicleID}: vehicle{as vehicles}. GET /vehicles/id/{vehicleID}/breaks: breaks[break_api_id, vehicle_api_id, window_start, window_end, break_timing, duration, created_at]. A name in fields is one of these, or one key below it and no deeper.

dispatch_invoices

ChatGPT
The invoice data on the jobs scheduled in a period of the user's linked Dispatch account: each job's invoices, and each of its actions' invoice_number, value and quantity. The dates are YYYY-MM-DD, the days the jobs are to be fulfilled, and Dispatch searches at most 30 days from start_date to end_date, so July 1 to July 31 is one call and a quarter is three. A call searches the period and reads up to 50 of its jobs, and may take 30 seconds. ok is true when the call read. period_jobs is how many jobs the period has and jobs the ones this call read, each with its job_id, date, invoices and actions, and each action with its invoice_number, value and quantity. totals_so_far counts every job read since the first call: jobs, jobs_with_invoice (those with a number in invoices), actions, actions_with_invoice (those with an invoice_number) and value, the sum of the actions' value. reads counts the call's reads of Dispatch. While the answer has next, call again with the same dates and cursor set to it, after retry_after_seconds when that is there; wait says why to wait. With no next and totals_so_far.jobs equal to period_jobs, every job was read. not_read has the jobs Dispatch refused or did not answer in two calls, each with its job_id, status_code and error: they are in no total, so say which are missing. Dispatch has no date and no total of an invoice, and does not say whether value is a unit price or a line amount.

dispatch_preview_route

ChatGPT
Preview a route of the user's linked Dispatch account without saving it: its stops in the order Dispatch would route them, its metrics and its warnings (POST /itineraries/action/previewRoute). Nothing is created or changed. route has the date, start_time, vehicle_id, start_point and end_point, and the job_ids or stops to route, as its schema says. fields keeps only the keys it names, offset and limit return part of each list, and an answer over 40,000 characters is refused with what to pass instead. Dispatch answers 403 when the account does not have route previews enabled. Keys of the answer, names only, none sure to be there, statusCode and status left out ([ ] a list's records' keys, { } an object's, * not in the contract): POST /itineraries/action/previewRoute: itinerary{itinerary_id, date, time, time_frame_size, start_location, end_location, route_status_code, route_error_message, round_trip, route_metrics, route_options, vehicle, operators, stops_count, stops, warnings}. A name in fields is one of these, or one key below it and no deeper.

dispatch_remove_from_route

ChatGPT
Take jobs or drivers off one route, an itinerary, of the user's linked Dispatch account: 1 to 50 a call. It sends DELETE /itineraries/id/{itineraryID}/job_id/{jobID} once for each ID of job_ids, or DELETE /itineraries/id/{itineraryID}/action/unassignOperators/operator_ids/{operatorID} once for each ID of operator_ids. Pass the route's itinerary_id and one of job_ids, the jobs to take off it, or operator_ids, the drivers to unassign, each once. A job comes off only while its stop has not started, and a pickup or dropoff stop comes off only with all of its related jobs. Dispatch deletes a route its last stop leaves. A removal has no undo: remove only what the user asked for. One removal answers as Dispatch did, with read_back, verified once the route reads back without the job. Several answer results in the order passed, each with its ok, status_code and read_back; a result's place in results is its ID's in the list, except one skipped after the call stopped, which names its entry. The removals go one after another, and those after one whose outcome is unknown are skipped. With stop_at_first_failure the first not accepted ends the call and skips the rest. The IDs not sent, after 30 seconds, when the answer is full or when the account's turn or minute refuses one, are left with why_left: call again with call_again_with, exactly as it is, which names the route and the list. ok is true when every removal was accepted.

dispatch_save_location

ChatGPT
Create a location, such as a depot, in the user's linked Dispatch account, or change one: one a call. It sends POST /locations for new_location, or PATCH /locations/id/{locationID} for location_id and changes. Pass new_location, a location as the contract declares it, or location_id and changes, the changes to that location; active false takes it out of use. A sub-user may not create or change one. The change is made at once with no undo: change only what the user asked for. The answer is Dispatch's, with the new location_id after a create, and read_back says whether it reads back as sent.

dispatch_save_vehicle_breaks

ChatGPT
Add a break window to a vehicle of the user's linked Dispatch account, or change one: one break a call. It sends POST /vehicles/id/{vehicleID}/breaks for new_break, or PATCH /vehicles/id/{vehicleID}/breaks/{breakID} for break_id and break_changes. Pass the vehicle_id and one of new_break, a break as the contract declares it, or break_id and break_changes, the changes to one of its breaks. A vehicle has up to 3 break windows, and a sub-user may not change them; an account whose plan has no breaks is answered 403. The change is made at once with no undo: change only what the user asked for. The answer is Dispatch's, with the new break's ID after an add.

dispatch_search_jobs

ChatGPT
Search the jobs of the user's linked Dispatch account by date and other criteria, 1 to 50 searches in a call: POST /jobs/search for each. Each search has start_date as YYYY-MM-DD, an end_date at most 30 days after it, and the criteria its schema lists, such as status, invoice_number or on_itinerary; a longer period is several searches in one call, of 30 days or less each. Dispatch answers the jobs' IDs alone, in in_staging.ids and post_staging.ids: read the jobs with dispatch_get_jobs, up to 50 a call. For one search the answer is the read's data. For more, results has one result for each search made, in the order given, with its search and its own ok, and left the searches still to make, as they were passed: pass them again, after retry_after_seconds when it is there. Many IDs are read in parts with fields ["in_staging.ids", "post_staging.ids"] and offset and limit; lists gives each list's total, in the answer for one search and in each search's result for more. An answer over 40,000 characters is refused with what to pass instead. An account reads 600 times a minute, 4 at once; a call past that says when to call again. Keys of the answer, names only, none sure to be there, statusCode and status left out ([ ] a list's records' keys, { } an object's, * not in the contract): POST /jobs/search: in_staging{count, ids}, post_staging{as in_staging}. A name in fields is one of these, or one key below it and no deeper.

dispatch_send_job_message

ChatGPT
Write a message in a job's Notes and Chat, in the user's linked Dispatch account, or edit one: one a call. It sends POST /jobs/id/{jobID}/chat for new_message, or PUT /jobs/id/{jobID}/chat/{messageID} for message_id and edited_message. Pass the job_id and one of new_message, or message_id and edited_message, the message's new text. Only a message created through the API can be edited, and an edited message shows as unread and notifies the job's operators. A message is sent at once with no undo: write only what the user asked for. The answer is Dispatch's; no read judges a message, so read_back is not_covered.

dispatch_update_driver

ChatGPT
Change a driver, an operator, of the user's linked Dispatch account: one a call. It sends PATCH /operators/id/{operatorID}. Pass the operator_id and changes, the fields to change as the contract declares them; active false blocks the driver's access. A driver's login, email and password, is changed in the Cigo dashboard and refused here. A driver is created in the Cigo dashboard, which sets their login (email and password); this server cannot create one. A sub-user may not change a driver. The change is made at once with no undo: change only what the user asked for. The answer is Dispatch's, and read_back says whether the driver reads back as sent.

dispatch_update_jobs

ChatGPT
Update jobs of the user's linked Dispatch account, or add or edit one job's action. It sends PATCH /jobs/id/{jobID} once for each entry of updates, POST /jobs/id/{jobID}/actions for new_action, or PATCH /jobs/id/{jobID}/actions/{actionID} for action_changes. Pass one of: updates, 1 to 50 entries, each a job_id a read returned and the changes to make to that job, as the contract declares them; job_id and new_action, an action to add to that job; or job_id, action_id and action_changes, the changes to one of its actions. A job on a route whose state is new takes changes only to quick_desc, its contact fields, apartment, time_preference, invoices and balance_owed. A job on a route takes no new pick up, return or exchange action, and a fulfilled job or action takes no action change. A job with an email or a mobile_number may be notified to its customer by Email and SMS. Each change is made at once with no undo: send only what the user asked for. One change answers as Dispatch did, with read_back. Several updates answer results in their order, each with its ok, status_code, read_back and the ids Dispatch named; a result's place in results is its entry's in updates, except one skipped after the call stopped, which names its entry. Two entries for one job go in order, and one after an entry for the same job whose outcome is unknown is skipped. With stop_at_first_failure the first not accepted ends the call and skips the rest. The entries not sent, after 30 seconds, when the answer is full or when the account's turn or minute refuses one, are left with why_left, to pass again as updates. ok is true when every change was accepted.

dispatch_update_route

ChatGPT
Change one route, an itinerary, of the user's linked Dispatch account: one change a call. It sends PATCH /itineraries/id/{itineraryID} for changes, PATCH /itineraries/id/{itineraryID}/action/updateOperators for crew, PATCH /itineraries/id/{itineraryID}/action/updatePositions for stop_order, PATCH /itineraries/id/{itineraryID}/action/reversePositions for reverse, or PATCH /itineraries/id/{itineraryID}/job_id/{jobID}/position/{position} for job_id and position. Pass the route's itinerary_id and one of: changes, its fields and the jobs it adds as add_ids; crew, its drivers, which replace those it has, at most 4 unless the account allows more; stop_order, its stops in their new order; reverse true, its stops in reverse; or job_id and position, one stop moved to that position. A route takes changes only while it is not completed, and a sub-user may not change one. changes takes no delete_ids: Dispatch deletes a route that taking jobs off empties, so jobs come off a route with dispatch_remove_from_route. Each change is made at once with no undo: change only what the user asked for. The answer is Dispatch's, with its warnings, such as jobs it did not add. read_back is verified when the route reads back holding the jobs added or the crew sent, and not_covered for other fields, an order or a position, which no read judges.

export_status

ChatGPT
How far an export of concluded_jobs_report got: status waiting, reading, done or failed with error, rows_written of rows_counted, and its download link. An export is kept 7 days.

provider_profile

ChatGPT
Read the signed-in Cigo user's provider profile, as Foresight's own chat reads it: the provider's usual job areas, its recent operators and fleet, and the dates they were observed over. It is background evidence, not the account's current state. provider_id is needed only when the user has more than one provider; another provider's profile is never returned.

resolve_area

ChatGPT
Match the words the user used for an area, such as "usual Laval area" or "the north shore", to the areas of the signed-in user's provider profile, as Foresight's own chat does. status says how it matched, and matches has each area with its center, radius in km and observed dates; none matched is ok: false. provider_id is needed only when the user has more than one provider.

sql_read

ChatGPT
Run one MySQL SELECT over Foresight's reporting tables as the signed-in Cigo user, and return its rows as JSON objects. The tables and columns are the ones contracts/sql/reporting-schema.txt declares: read it with workspace_read. What a word of the business or a metric means in these tables, such as which jobs "deliveries" or "completed" count and which date column a period uses, is defined in the pages knowledge/foresight/index.md lists: workspace_search finds a word in them. A row's api_id is the ID the Dispatch tools take, and its id the number on its page in Cigo: tbl_package for a job, tbl_schedule for a route. Every table is read through the user's providers, and a merchant's through its locations, as Foresight's own chat reads; another provider's rows are never returned. A job still in staging, the Import Tool, may be missing from these tables: a count or list of jobs that should include them adds what dispatch_search_jobs answers under in_staging when this connection lists that tool, and otherwise says those jobs may be missing; the answer says which jobs it counts. One statement, no lock, variable or file; a refused one is answered with the reason. rows is how many to return, 200 unless you pass it and at most 1,000, within an answer of 40,000 characters. truncated says when the rows are not the whole result: aggregate in SQL, or page with ORDER BY, LIMIT and OFFSET. A user makes 60 SQL reads a minute.

workspace_list

ChatGPT
List Cigo Foresight's reference files, each with its path, lines and characters. They are what Cigo's words and metrics mean, its reporting tables and how they join, its reports and the click paths of its screens (knowledge/foresight/, which index.md there lists by subject); the tables and columns of the reporting data and Dispatch's API contract (contracts/); and notes on Dispatch's API (.agents/skills/cigo-dispatch/references/). under keeps one folder, such as knowledge/foresight/policies. Read a file with workspace_read, and find text in the files with workspace_search.

workspace_read

ChatGPT
Read one of Cigo Foresight's reference files by its path from the top, as workspace_list writes it: knowledge/foresight/index.md. A link inside a file is relative to that file's folder. The text comes in whole lines from line, 1 the first, and limit is how many. An answer holds 40,000 characters: lines says the file's total, the first line returned and how many came, and when the file goes on, note gives the line to call again with. The files describe Cigo's data and screens, and were written for Foresight AI, the assistant inside Cigo's own application. Where one speaks of that assistant's page, such as a table cut at 25 rows, Download CSV or a link like /job/<id>, or names its AGENTS.md or a skill, it means Cigo's application: this server has no such page and serves neither file.

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/cigo-foresight"><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/cigo-foresight)