MCP Server

Ask your AI assistant. Get answers from your own papers.

You already write with Claude, Cursor, or another AI assistant. Connect it to Agent Bayes, and instead of answering from memory it searches the papers you indexed, then comes back with real passages, real page numbers, and references you can put straight into a bibliography.

How it works

How can my assistant read my library?

Through MCP. It's a standard plug that lets AI assistants use outside services, and Agent Bayes now speaks it. You create one key in your Agent Bayes settings and paste it into your assistant's configuration. From then on, when you ask a question about your research, the assistant can search your indexed papers, open your mindmaps, and hand the hard questions to the Agent Bayes research agent. Nothing to export. Nothing to copy over.

Citations that survive the conversation

Every passage your assistant retrieves arrives with its page range and the paper's full reference attached. When an answer makes it into your draft, the citation is already there. And it's real.

You decide who does the thinking

Want to stay hands-on? Have your assistant search your papers, then compose from the passages yourself. Facing a harder question? Hand it to the Agent Bayes research agent, which answers from your sources and shows which passages it used. Both paths are built in.

Your assistant reads the manual first

Before it can use any tool, your assistant must fetch the Agent Bayes operating manual. The manual teaches it the citation rules, the current credit prices, and to try the cheapest model first. You don't have to explain anything.

The whole workspace, not a window into it

Create projects and knowledge bases. Upload and index new PDFs. Manage tags, read and edit mindmaps, continue earlier agent conversations. Everything runs under the same limits as the app, with a key you can revoke in one click.

Installation

Connected in three steps

1. Create a key
In the app, open Settings → API Keys and click Create API key. The key appears once. Copy it now. If you use more than one device, make a key for each, so you can revoke one without breaking the others.

2. Tell your assistant about it

Using Claude Code? One command:

claude mcp add --transport http agent-bayes \
  https://api.agentbayes.com/mcp/ \
  --header "X-API-Key: sk-your-key"

Using Cursor or another app that takes a JSON config? Add this:

{
  "mcpServers": {
    "agent-bayes": {
      "url": "https://api.agentbayes.com/mcp/",
      "headers": { "X-API-Key": "sk-your-key" }
    }
  }
}

The same key also works as Authorization: Bearer sk-your-key if your client prefers standard bearer headers.

3. Ask something
Try: "Look at my Agent Bayes projects and tell me what my papers say about X, with citations." Your assistant fetches the operating manual, then gets to work.

Three things to know first

Your key is your account. Anyone holding it can do what you can do. Treat it like a password. If it leaks, revoke it in Settings and the connection dies immediately.
Reading is free. Spending is announced. Searching and reading cost nothing. The operations that spend credits, like indexing a PDF or running the research agent, say so before your assistant uses them, and they settle to what was actually used.
You need an Agent Bayes account. Same plans, same limits, same credits as the app. Your assistant sees what your account can see. Nothing more.

Ready to connect

Your library, in every conversation

Index your papers once. Then every assistant you use knows where to look.

Start now Read the docs