Plug your brain into any LLM client

Sign in with your Rapu API key. We'll generate copy-paste setup for Claude Code, Claude Desktop, and raw MCP — anywhere you can run an LLM, you can run it against your brain.

Couldn't sign in with that key. Check it and try again.
Don't have a key yet? Set up your brain in the app first.
LLM setup
Workshop → Chat →

Connect Rapu to your LLM client

Rapu is your brain. Any MCP-capable LLM client can talk to it — pulling claims as grounding, scoring predictions, walking argumentation graphs, simulating retractions. Pick your client below; the setup snippet is generated with your API key already filled in.

Before you connect, read what this actually means. Your LLM provider sees both your prompt and any atom content the tools return. The LLM is not "trained" on your brain — atoms are passed as context per query. "Calibration" tracks your record, not whether your claims are true. Full honest version →
MCP URL
API key
Primary brain

1. Add the MCP server

In your terminal, in any directory (this is a per-user config, not per-project):

2. Verify the connection

Open Claude Code and try one of the starter prompts below. If you see Rapu's tools in the tool list (or the LLM uses them), you're connected.

To remove the connection later: claude mcp remove rapu

Heads up: pasting the MCP URL into a Claude chat does nothing — it's just text. Claude Desktop has to be told about Rapu as a server. There are two ways; pick one.

Option A — Config file (works with your API key, needs Node.js)

Claude Desktop's config only runs local servers, so a small bridge (mcp-remote) connects it to Rapu over HTTP. You need Node.js installed (it provides npx).

1. Open the config file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

2. Merge this in (alongside any existing mcpServers):

3. Fully quit and reopen Claude Desktop. The Rapu tools should appear in the tools (🔌) panel.

Option B — Connectors UI (no terminal, sign-in based)

In Claude Desktop: Settings → Connectors → Add custom connector. Paste the MCP URL:

Claude opens a Rapu authorization page — paste your Rapu API key once to approve, and Claude stores its own connector token (revoke it anytime under your API keys). No config file, no terminal. Note: Claude's custom connectors require a paid Claude plan (Pro/Max/Team/Enterprise). Once authorized, the Rapu tools appear in chat.

For SDKs, custom agents, or any HTTP-aware LLM loop

Rapu's MCP server speaks JSON-RPC over HTTP. Authenticate with a bearer token; call methods like tools/list and tools/call.

Example — list available tools:

Call a tool

For the full MCP spec see modelcontextprotocol.io. Rapu implements the standard server protocol — anything that speaks MCP can connect.

Starter prompts

Try these in your connected client — they exercise the apparatus end-to-end. Click to copy.

  • → "List my past-due predictions, summarize what each commits to, and suggest what evidence I'd need to score them." copy
  • → "Walk the argumentation graph on the fundraise topic. Which claims are accepted under grounded extension, and what would happen if I retracted the family-office hypothesis?" copy
  • → "Surface atoms past their decay half-life. For each, suggest whether to re-validate or supersede, and draft replacement wording." copy
  • → "Render a commissive projection for our Q4 product launch. Cite relevant existing atoms; flag what's missing." copy
  • → "Show me where my brains agree vs disagree about ESG framework requirements. Highlight cross-standpoint divergence." copy
Once you've connected, the LLM has access to ~30 tools covering atoms, projections, predictions, calibration, argumentation, brain links, and counterfactual simulation. Ask it what it can do — it'll tell you.