MCP App Store
by KeenEthics
Shruwd icon

Shruwd MCP Server for ChatGPT

By Shruwd

Overview

Shruwd shows how often your brand appears in ChatGPT and Google AI Overviews answers, which competitors appear instead, and which sources get cited. It explains why you're missing, recommends a specific fix, and after you ship it, checks again to show whether it worked. From ChatGPT you can: - Check your visibility, share of voice and citation share against competitors, and how they're trending - Read the actual AI answers behind every number - See what's holding your brand back, with the recommended fix for each finding - Set up a brand: add prompts and competitors, or have Shruwd suggest them from your homepage - See which AI crawlers reach your site Connect with your Shruwd account. A free account includes one measurement of your domain.

How to connect

  1. ChatGPT: Open the Shruwd listing in the ChatGPT app directory and choose Connect. ChatGPT listing Shruwd exposes 31 tools here.

Tools

shruwd_accept_suggestion

ChatGPT
Turns a suggestion into a tracked competitor, with exactly the rules of shruwd_add_competitor: exact aliases only, and a short or common-word name or alias is refused until contextTerms are given. The name and the linked domain are prefilled from the suggestion; pass aliases, domains, exclusions or contextTerms to adjust. Confirm with the user before accepting.

shruwd_add_competitor

ChatGPT
Adds a competitor. Matching is exact, on word boundaries — no fuzzy matching, ever — because a false positive silently corrupts every metric. That is also why a short name or alias (six characters or fewer) or a common English word ("Arc", "Linear", "Notion") is refused without contextTerms: those names appear in sentences that are not about the company. Give one or two category words as context terms ("project management", "crm") and the mention counts only when one appears nearby. Include the competitor's domain so citations of it are attributed. On a new brand, add competitors before any prompt: the first measurement starts as soon as prompts exist, and a competitor added after it is not in it.

shruwd_add_prompts

ChatGPT
Adds prompts — the questions asked of each engine every cycle. Write them the way a buyer would type them, without the brand name unless the intent is navigational. 1–500 characters each. The batch is atomic: if it would exceed the plan's prompt allowance — one pool shared across every brand in the workspace, not a per-brand limit — nothing is added and the error carries limit, current and submitted. On a new brand the first measurement starts within minutes of the first prompts and covers only what exists then, so add its competitors before this and every prompt in one call.

shruwd_archive_brand

ChatGPT
Archives a brand. Nothing is deleted; scheduling stops and history stays. Requires the workspace owner: a key held by a member is refused with 403 insufficient_role. Ask before using this.

shruwd_create_brand

ChatGPT
Creates a brand, its self entity (one alias = the name, one domain = the registrable domain) and the first measurement cycle. If the account has no workspace yet, one is created on the free tier. The response says which engines the plan measures and whether a first cycle was planted. On the free tier "snapshot" is "planned" or "already_taken": a domain gets one free measurement ever, across all accounts. Add competitors first (shruwd_add_competitor), then every prompt in one shruwd_add_prompts call: the first cycle starts within minutes of the first prompts and measures only the prompts and competitors that exist then. The name is the brand's own first alias, so a short (six characters or fewer) or common-word name is refused with context_terms_required until contextTerms are given, as for a competitor. Requires the workspace owner: a key held by a member is refused with 403 insufficient_role.

shruwd_create_ingest_token

ChatGPT
Mints the token the brand's server sends its access logs with, and returns the endpoint to POST them to plus a setup object with three routes: an HTTP log drain (Vercel, Netlify, Fly), a Cloudflare Logpush HTTP destination (Enterprise zones; the destinationUrl embeds the token), or the reference forwarder Worker (any other Cloudflare plan or CDN-served site). The token is shown once and never again; previous tokens are revoked. Ask which platform the site runs on, then give the user that route's values verbatim. Logs power the crawler view and the highest-confidence findings. Requires the admin role on the brand: a key held by an editor or viewer there is refused with 403 insufficient_role.

shruwd_dismiss_suggestion

ChatGPT
Not a competitor. Quiet for ninety days; suggested again only if the answers still name it.

shruwd_get_brand

ChatGPT
One brand with its pipeline health: last successful cycle, next scheduled cycle, run success rate, log-ingest continuity and connection states. Check this before treating a flat metric as a signal — "nothing ran" and "nothing was mentioned" look identical in a chart.

shruwd_get_crawlers

ChatGPT
Which AI crawlers fetched the brand's site, from its server logs, over a window. verifiedHits were confirmed against the vendor's published IP ranges; unverifiedHits merely claimed the user-agent, which anyone can send. Only verified hits are a signal, and the two are never summed. "live_retrieval" bots fetch pages to answer a question now, so a page they cannot fetch cannot be cited in those answers; "training" crawlers collect pages for model training. "coverage" says whether log ingest has been continuous — if not, an absence of hits means nothing.

shruwd_get_finding

ChatGPT
One finding with its full evidence, recommendation, and the history of state changes. A finding the plan does not show in full — one counted in a findings list's "locked" — is refused with 403 not_entitled: it exists, the plan withholds it.

shruwd_get_visibility

ChatGPT
Mention rate and share of voice for the brand and each competitor on one engine over a window (default: last 30 days), plus per-prompt coverage. Each metric is {state: "ok", point, lo, hi, n} — a proportion with its 95% interval — or {state: "insufficient_data"} when fewer than ten responses exist, or {state: "undefined"}. A metric with state insufficient_data is NOT zero; do not report it as a number, and do not compare two metrics whose intervals overlap as if one were higher. "asOf" is the day the numbers describe. "historyFrom" is set when the window was clamped to the plan's history. "modelIds" lists the provider models seen; a change there can move metrics on its own. engine defaults to google_aio: call once per engine the brand measures (its engines) and report each separately.

shruwd_get_visibility_series

ChatGPT
The window metric at every cycle close in a range (default: the last ninety days, 30-day window): for the brand and each competitor, mention rate and share of voice per close, each a {state: "ok", point, lo, hi, n} metric or an explicit non-number. Use it to answer "is this changing" — but two points whose intervals overlap have NOT moved, and a change across a "modelChanges" entry may be the provider's model rather than the brand (the comparison is confounded). "series" is index-aligned with "points"; null means no rollup row for that close.

shruwd_get_workspace

ChatGPT
The plan, the entitlements in force (brands, the workspace-wide prompt pool, members, engines, cadence, history window, rechecks per month, and maxVisibleFindings — how many findings the plan shows in full), usage against the period allowance including activePrompts against promptsAllowance, whether the caller is the workspace owner or a member, and the brands the caller can reach with their role on each. Call this first: the role on a brand decides which of these tools are allowed on it. Usage is the whole workspace's, since the pools are shared.

shruwd_list_answers

ChatGPT
The latest individual AI answers for the brand, newest first: the prompt, the engine, whether the brand was named and at what rank, the competitors named and the pages cited. Evidence, not a metric: never compute a rate from these; use shruwd_get_visibility, which withholds numbers below ten answers. To read the answers behind a finding on a prompt, pass its subjectRef as promptGroupId.

shruwd_list_brands

ChatGPT
The active brands the caller can reach, with their prompt and competitor counts and the caller's role on each.

shruwd_list_crawler_pages

ChatGPT
Which pages of the brand's site AI crawlers fetched most, every bot on a page added up, verified hits only. liveRetrievalHits are fetches for answers being written now (a ChatGPT-User fetch means someone's question pulled that page); trainingHits are crawls for model training, which tend to sweep every page evenly, so they say little about which pages matter. Ordered by live retrieval, then training; purpose "training" orders by training instead. bots says which bots fetched each page. Pass contentOnly to drop files, framework data and sign-in pages. A page fetched is not a page cited: shruwd_list_sources says what the answers cite. For one bot, or where errors came from, use shruwd_list_crawler_paths. Kept 90 days.

shruwd_list_crawler_paths

ChatGPT
Which paths each AI crawler fetched and what it got, one row per bot and path, verified hits only. Use status "error" to find where the errors in shruwd_get_crawlers came from. Read contentPage before calling an error a problem: it is false for files and framework data (_payload.json, scripts, images, feeds) and for sign-in and account pages, and a framework data file can 404 to everyone by design. worstStatus tells a missing page (404) from a refusal (403). totalPaths above the rows returned means the list was cut. Per-path detail is kept 90 days.

shruwd_list_cycles

ChatGPT
Recent measurement cycles, newest first, with state (pending, running, complete, partial, failed, skipped_quota) and run counts. "partial" means some runs failed; "skipped_quota" means the allowance, the plan or billing stopped it. On a weekly plan the newest is next week's cycle, still pending: the newest that is not pending is the current measurement, and its results are in once it is complete or partial.

shruwd_list_entities

ChatGPT
The brand itself (isSelf) and every competitor, each with the exact aliases, domains, exclusions and context terms that decide what counts as a mention or a citation.

shruwd_list_findings

ChatGPT
Diagnoses, ordered by severity, then confidence, then estimated impact. Each has a specific recommendation: what to change, where, and which metric should move. confidence: "observed" was measured from first-party data; "inferred" follows a reliable pattern; "heuristic" is a plausible cause that could not be confirmed — treat it as a hypothesis, not a fact. By default the open, acknowledged, fix_applied and rechecking findings; pass states to change that. "queued" counts open findings held back by the weekly cap. On plans that cap visible findings, "locked" counts the rest by severity: they exist and are not returned. When locked.count is above zero, never present the returned findings as the complete set — say how many are withheld.

shruwd_list_prompts

ChatGPT
The brand's prompts. With an engine, each prompt carries its coverage over the window: how often the brand was mentioned in answers to it. Per-prompt n is small (three repetitions per cycle), so coverage is often insufficient_data; that is correct, not missing data.

shruwd_list_sources

ChatGPT
Which sites and pages the AI answers cite, and whose they are: one row per domain, most-cited first, with its most-cited pages. Use it to see what the engines read for the brand's questions, or for one question (pass a promptGroupId, or a finding's subjectRef). kind says whose a domain is: yours, a competitor's, or a third party (review_directory, forum, wikipedia, video, editorial). rate is the share of answers citing it with its 95% interval; state insufficient_data is not zero, never report it as a number. answers.citingAny is how many answers cite anything at all: where few do (ChatGPT often cites nothing), a share means less. yourAnswers counts answers citing your own pages there, or videos on your named YouTube channel. findings lists the live findings about a source.

shruwd_list_suggestions

ChatGPT
Names the AI answers use that the brand does not track yet — the products a person would recognise as competitors but never added. Each comes with the share of answers that named it (e.g. 0.34 = in 34% of a prompt cluster's responses) and, when the answers linked it, its domain. Review them with the user: accept the real competitors (they then count in share of voice), dismiss the rest. Suggestions appear only after a measurement cycle has closed.

shruwd_remove_entity

ChatGPT
Closes a competitor as of now. The brand's own entity cannot be removed.

shruwd_remove_prompt

ChatGPT
Deactivates a prompt. Nothing is deleted; its history stays readable.

shruwd_run_measurement

ChatGPT
Schedules a measurement cycle now: every active prompt, on every entitled engine, repeated as the plan sets (three times on weekly plans, once for the free snapshot). It starts at once and completes over the following day; this response contains only the cycle id. Watch it with shruwd_list_cycles, then read shruwd_get_visibility and shruwd_list_findings. A cycle spends the period's run allowance (see shruwd_get_workspace); weekly plans already run one per week, so use this after a change worth measuring. Do not call it to start a new brand: shruwd_create_brand plants the first measurement, and it starts by itself once the brand has prompts. On the free tier that first measurement is the single free one.

shruwd_set_entity

ChatGPT
Sets the DESIRED aliases, domains, exclusions and context terms of an entity (the brand's own self entity included). Send the full lists: what is missing is closed as of now, what is new is added, what is unchanged is untouched, and past measurements keep the aliases that were valid when they ran. Read shruwd_list_entities first.

shruwd_suggest_setup

ChatGPT
Reads the brand's homepage and drafts ten prompts and up to six competitors. Nothing is saved to the brand: show the draft to the user, then add each competitor they confirm with shruwd_add_competitor, passing its aliases and contextTerms from the draft, and after that the prompts they keep, in one shruwd_add_prompts call. Returns the stored draft when one exists unless regenerate is true. Three drafts per brand per day; a new draft can take up to a minute.

shruwd_transition_finding

ChatGPT
Moves a finding: "acknowledged" (seen), "fix_applied" (the recommended change is live), or "dismissed". Order matters: an open finding must be acknowledged before fix_applied, or the call is refused with 409 invalid_transition. fix_applied captures a baseline of the affected prompts and starts a fourteen-day clock; a recheck then measures again, and the finding becomes "resolved" ONLY if the movement gates pass (non-overlapping intervals and at least five points of change), else "not_moved". ACCESS findings are instead confirmed from the site's own logs, robots.txt or page, with no wait and no recheck spent. Neither resolved nor not_moved can be set here. Do not mark fix_applied until the change is actually deployed; the baseline is taken at that moment. A finding withheld by the plan is refused with 403 not_entitled.

shruwd_update_brand

ChatGPT
Renames a brand or changes its timezone. The name is for display; matching uses the self entity's aliases (shruwd_set_entity). The timezone moves the day boundary for cycles and rollups from the next cycle on. The domain cannot change: it defines the self entity, so a new domain is a new brand.

shruwd_update_prompt

ChatGPT
Edits a prompt by its promptGroupId. Changing text or intent creates a new version under the same group — history stays attached to the version that produced it. tags, active and page change in place. Reactivating counts against the plan limit. page is the page on the brand's own site that should answer the question: when a finding on a prompt has recommendation.target null, or names the wrong page, set it here (the finding's subjectRef is the promptGroupId). Findings name it from the next measurement, not at once.

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/shruwd"><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/shruwd)