Start with a link
If your assistant can open web links, copy the address of a contribution page into your chat and ask a question. You can try this prompt without setting up a connection:
Read https://aiproofwatch.com/archive and find an AI contribution in a topic I am interested in. Explain what was claimed, why it matters, and what has been checked. Include a link to the record.
Add AI Proof Watch to your assistant
You can also add a connection so your assistant can look up topics and problems directly. This uses MCP, a standard way for AI assistants to get information from other services.
Your assistant needs to support custom MCP connections and be able to open web pages. If you cannot find a custom connection option, check your app’s help or use the link method above.
- Open your assistant’s settings and find its connections or integrations.
- Add a custom connection or MCP server. Use AI Proof Watch as the name and paste the address below.
- Save the connection and enable it in your chat if asked. You do not need an AI Proof Watch account, API key, or login.
https://aiproofwatch.com/mcpMenu names vary by app. These are general steps; setup in individual assistant apps has not yet been tested.
Technical setup and connection tests
If your app asks for a connection type, choose Streamable HTTP. Choose no authentication if asked. The connection has been tested with MCP Inspector’s command line and the official TypeScript SDK, including legacy protocol support.
Try MCP Inspector
With Node.js and npm installed, run:
npx @modelcontextprotocol/inspector --cli https://aiproofwatch.com/mcp --transport http --method tools/listThis lists the two discovery tools. The Inspector command runs without an AI model.
Use the TypeScript SDK
Install @modelcontextprotocol/client@2.4.0, then connect:
import {Client, StreamableHTTPClientTransport}
from '@modelcontextprotocol/client';
const client = new Client({name: 'research-reader', version: '1'});
await client.connect(new StreamableHTTPClientTransport(
new URL('https://aiproofwatch.com/mcp')
));
const result = await client.callTool({
name: 'search_interests',
arguments: {query: 'combinatorics', limit: 10}
});
console.log(result);
await client.close();Public data
Agents and feed tools can read the public JSON manifest. It links to reviewed records, search indexes and public change history.
Try a question
The connection finds topics and problems and provides their background. It links your assistant to contribution pages for the claims, ratings, checking evidence and revision history.
Find recent AI contributions in combinatorics. Explain why they matter and include links to the sources.
What exactly was checked for this result, and by whom?
Explain why this result matters, assuming undergraduate mathematics.
Compare these contributions by AI involvement, verification, and human understanding.
Ask your assistant to keep the sources, assumptions and uncertainty in its answer. Its explanation is generated by the assistant, not reviewed by AI Proof Watch.
Want regular updates?
For updates in your inbox, choose email updates. You do not need to connect an AI assistant.
For a regular briefing in your assistant, check whether it supports scheduled tasks and notifications. Adding this connection does not turn them on. To answer “What changed since my last check?”, your assistant also needs to remember what it read before.