What the Promptwatch MCP Server Can Actually Do: Read Tools, Write Tools, and Scopes
A catalog of the real read and write tools the Promptwatch MCP server exposes, grouped by job, with the scope boundary that keeps agents safe.
An MCP server is only as useful as the tools it hands an agent. A long list of vague names is worse than a short list of precise ones, because the agent has to guess what each call does. This post catalogs the real tools the Promptwatch MCP server exposes at https://server.promptwatch.com/mcp, grouped by the job each tool does, and then covers the scope boundary that decides which tools an agent is even allowed to see.
The names below are the actual tool names, not paraphrases. If you are wiring Claude Code, Cursor, or any MCP client to Promptwatch, these are the calls your agent will make.
Prompts and visibility
The prompt is the unit of work in AI search. These tools let an agent read what you track and how it moves.
listPrompts returns the tracked prompt set for a project. getPrompt reads a single prompt in detail, including the current answer and the engines it runs against. getVisibilityTimeSeries shows how a prompt's visibility and rankings moved over time, which is the first call in any diagnosis. listQueryFanouts exposes the fan-out structure around a seed query, the related prompts an answer tends to pull in, which is how you find the real query behind a tracked one.
Citations and competitors
Citations are where AI systems source their answers. These tools let an agent see who is being cited and for what.
getCitations returns the citation set for a prompt or project. getCitationTopPages narrows that to the pages that get cited most, which is the evidence you need before proposing a content fix. listRedditCitations and listYoutubeCitations pull the two off-domain channels that matter most for reach, since a lot of AI answers quote Reddit threads and YouTube transcripts. getCompetitorHeatmap shows who shows up alongside you across prompts, which tells you whether a loss is yours alone or a category shift.
Content gaps and site health
The acting half starts with finding the work. These tools surface what is missing.
listContentGapPrompts returns the prompts where you have no coverage. getContentGapRecommendations turns those gaps into suggested work, with the reasoning attached. getContentGapStats gives the counts and shape of the gap so an agent can prioritize. getSiteHealth reports structural problems that could block crawlers or indexing, which is the evidence a technical fix needs before anyone writes a robots.txt change.
Crawler logs and visitor analytics
This is the surface that separates a visibility platform from a prompt tracker. These tools let an agent read AI crawler behavior and the traffic that follows.
getCrawlerTrend shows how AI crawlers hit your site over time. getTopCrawlerPages shows which pages they read most, which is the direct evidence for crawl-to-citation reasoning. getVisitorTrend, getVisitorTopPages, and getVisitorTopSources attribute the AI-referred traffic and conversions that follow a visibility change. An agent that reads both sides can tell a coherent story: a crawler hit this page, the page got cited for this prompt, and this much traffic followed.
Sentiment, ads, and shopping
getResponseSentimentTimeSeries tracks sentiment in the answers over time. listAds and listAdPrompts cover sponsored placements inside ChatGPT answers, the Ads Radar surface. listShoppingItems reads tracked shopping items, and getContentGapStats doubles as the prioritization input for shopping coverage. These are narrower surfaces, but they are the ones that let an agent reason about paid and shopping placement rather than only organic visibility.
Actions
listActionItems reads the action board, the queued work the platform already discovered. An agent that ends a diagnosis by reading the board can avoid duplicating work and can update existing items instead of creating new ones.
The write tools
The write surface is smaller, and that is the point. These are the calls that mutate state, and each one maps to a concrete acting step.
createPrompt and createPromptsBulk add tracked prompts, which widens the measurement set. createActionItem and updateActionItemStatus manage the action board, which is how a diagnosis becomes owned work. createReport builds a report from live data, which is the call that ends a weekly review loop. addShoppingTrackedProducts extends shopping tracking.
The content write tools drive the Content Agents workflow. createContent drafts from a gap or brief. listContentSlots reads the review inbox. acceptContentSlot moves a draft forward, and publishContentLive pushes the approved draft to the connected CMS. That last call is the boundary where work leaves the platform and reaches your site.
The scope boundary: read-only keys hide write tools
The single most important property of this server is that a read-only key hides the write tools from the client. When you connect an agent with a read-only key, the write tools are not present in the tool list. The agent literally cannot call publishContentLive or createContent, because the client never received them.
That is a stronger guarantee than a permission flag. A permission flag says the call is allowed or denied at runtime. A hidden tool says the call does not exist for this session. For an agent that runs on a schedule, or that you hand an open-ended goal, the hidden-tool model is the one you want.
The practical pattern is two keys. Keep a read-only key in your permanent config for exploratory and scheduled work, where the agent diagnoses, reads, and proposes. Keep a write key separate, and only add it to the config for the narrow loop you have reviewed and want to gate behind a human. The write key reveals createContent, listContentSlots, acceptContentSlot, publishContentLive, createReport, createActionItem, updateActionItemStatus, createPrompt, createPromptsBulk, and addShoppingTrackedProducts. Nothing else changes. The read surface is identical.
How to use this catalog
When you build an agent workflow, pick the smallest set of tools that closes the loop. A diagnosis loop needs listPrompts, getVisibilityTimeSeries, getPrompt, getCitations, getCitationTopPages, getCompetitorHeatmap, and listActionItems. A content loop adds listContentGapPrompts, getContentGapRecommendations, and, behind the write key, createContent, listContentSlots, acceptContentSlot, and publishContentLive. A crawler loop needs getCrawlerTrend, getTopCrawlerPages, and getSiteHealth. A report loop needs the read tools plus createReport.
The Promptwatch MCP server exposes a full measure, decide, and act surface, and the scope model lets you hand an agent the acting half only when you are ready. Start with the read tools, keep the key read-only, and widen the scope one loop at a time.