# How do I connect DecisionPen MCP to Cursor or Claude?

> Canonical: https://decisionpen.com/docs/connect-mcp-cursor-claude  
> GitHub wiki: https://github.com/DecisionPen/decisionpen-mcp/blob/main/CONNECT.md

## Summary

To connect DecisionPen to Cursor or Claude, create an Edge API token under Admin → Security & audit → Edge API tokens. Install the Context Engine in a dedicated Python virtualenv. Set `DP_API_BASE` and `DP_API_TOKEN`, then add the MCP server to Cursor or Claude Desktop. Start in `cite_only` mode so agents can read the plane. Switch to `propose` when you want them to publish updates for Inbox review. Nothing joins the trusted map until a person approves it.

## Steps

1. Create an Admin Edge API token and store it as `DP_API_TOKEN=dp_edge_…`.
2. Create a dedicated venv: `python -m venv .venv-mcp`, then `pip install -r decisionpen_mcp/requirements.txt`.
3. Add the server to your Cursor MCP config (use absolute paths on Windows).
4. Use the same shape in Claude Desktop MCP settings, then restart Claude.
5. Optional: set `DP_MCP_MODE=propose` and review proposals in Inbox.
6. Optional: connect S3 documents — https://decisionpen.com/docs/mcp-s3-document-store  

## FAQ

**Where is the token created?**  
Admin → Security & audit → Edge API tokens.

**Do I need an Anthropic API key for MCP tool calls?**  
No. The MCP server talks to DecisionPen with your Edge token.