AI assistant connection
Use ProseMap from an AI assistant
ProseMap has a connector that speaks the Model Context Protocol (MCP). An assistant connected to it can estimate, run and read ProseMap analyses on your account, within a spending limit you set. It uses the same account, saved results, credit and rules as the web app.
Set it up
The server address is the same for every assistant:
| Server URL | https://www.prosemap.io/api/mcp |
Sign in (the easy way)
- In your assistant, add a custom connector, or a remote MCP server, and give it the server URL above.
- The assistant opens ProseMap in your browser. Sign in, or create an account.
- Set a monthly spending limit for that assistant and press Allow. You go back to the assistant, connected.
This works in assistants that support signing in to connectors, including Claude (Settings, then Connectors, then Add custom connector), ChatGPT in developer mode, and Claude Code:
claude mcp add --transport http prosemap https://www.prosemap.io/api/mcp |
The connection appears in the app under Account, where you can see what it has spent and remove it.
With a token
For assistants and scripts that cannot open a browser:
- Sign in to the app, open Account, and under Connect an AI assistant create a connection. Give it a name and a monthly spending limit.
- Copy the token. It is shown once.
- Add the server URL to your assistant with the header
Authorization: Bearer <your token>.
claude mcp add --transport http prosemap https://www.prosemap.io/api/mcp --header "Authorization: Bearer <your token>" |
If your assistant only accepts a URL and cannot send a header, use https://www.prosemap.io/api/mcp/<your token> instead. Treat that URL as a password. The token method has been verified end to end with Claude Code.
What the assistant can do
| Tool | What it does | Cost |
|---|---|---|
| estimate_analysis | Says what analysing a text would cost, and whether a saved analysis already exists | Free |
| submit_analysis | Starts an analysis within a maximum you state | Same price as in the app |
| get_analysis | Progress, then scores, bands, ranked hotspots and a link to the full visual report | Free |
| get_sentences | Each sentence's score, band, role and flags, and its handoff to the next | Free |
| get_guidance | The revision guidance as Markdown: your draft with a comment at every sentence and handoff that needs work, plus your saved notes | Free |
| explain_sentence | Why a sentence scored as it did | Same price as in the app; free to reopen |
| compare_versions | Score changes between two analysed drafts | Free |
| list_analyses | Your saved analyses | Free |
| get_balance | Your credit and how much of the connection's limit is used | Free |
Working on a draft with an assistant
The scores are evidence, not the goal. A deliberate fragment or a hard turn can be right for the piece even when it costs a few points. So the default way of working is: find the problems, agree which ones matter, and change only those.
Three ways to ask, from lightest to heaviest:
| Diagnosis only | "Analyse this draft with ProseMap. Tell me the three weakest passages and why. Do not rewrite anything." |
| Revision I approve | "Analyse this draft with ProseMap. Show me the weakest passages and, for each, a proposed change and what it would cost me in voice or meaning. Change only the ones I approve, then analyse again and show me what moved." |
| Bounded experiment | "Analyse this draft with ProseMap, revise from the guidance and re-analyse, for at most two rounds and $3 in total. Keep anything I marked as deliberate. Stop when the problems we agreed on are fixed, and show me both versions so I can choose." |
How a round works. Each round is one analysis; everything else in it costs nothing.
estimate_analysis: see the price. Agree what you authorise and how many rounds.submit_analysis, thenget_analysisuntil it is complete.- The assistant shows you the weakest passages. You say which ones to change and which are deliberate.
get_guidance: the assistant revises only the passages you agreed on, as the comments say, and nothing else.submit_analysisagain with the revised text andrevision_ofset to the previous id. The result reports the change from the previous round.- Stop when the problems you agreed on are fixed and you accept the result. Stop too if a round does not help.
Two rounds is usually enough. Most good prose scores between 70 and 85, and a higher number is not always better writing. The same instructions are sent to the assistant when it connects.
Limits and safeguards
- Spending limit. Each connection has a monthly limit that ProseMap enforces itself. An assistant cannot raise it or spend past it, whatever it asks for.
- Nothing runs unpaid. Only two tools use a model:
submit_analysisandexplain_sentence. Each reserves its maximum from your balance before it starts and is refused if the balance cannot cover it. Every other tool is free and calls no model. - Per-analysis maximum. Every analysis is started with a stated maximum and is never charged more than that. A failed analysis costs nothing.
- Same rules as the app. Credit is spent in the same order, saved results are free to reopen, and complimentary access applies in both places.
- Not a general model. The connector only exposes ProseMap's own analyses. It does not answer arbitrary prompts.
- Size. Up to 400,000 characters per document. Long documents take a few minutes; the assistant polls for the result.
- Dictation, notes and disagreements are in the web app only. The assistant can open the full report by giving you its link.
- Revoking. Remove a connection in Account and it stops working immediately, however it was made.
ProseMap's scores and explanations are produced by AI models. They can be wrong, incomplete or misleading, and they are one reader's judgment rather than an objective measure of writing quality. Check anything that matters independently.
Last updated .