Memtro for Pi
An open-source extension, or a paragraph of JSON: the terminal coding agent on Memtro as its model, with Memtro's tools over MCP.
Pi is a minimal agent harness from Earendil Works that runs in the terminal. The pi-memtro extension registers both halves in one install — Memtro as a model provider and Memtro's MCP server for its tools — and discovers the organisation's model ladder from /v1/models rather than asking anybody to keep a list in step. Writing ~/.pi/agent/models.json and mcp.json by hand does the same thing. Either way the model id is the part that matters: Pi runs an agent loop of its own, so auto:chat pins one reply per request, where a plain auto on an organisation whose default is agent nests Memtro's investigation inside Pi's loop.
What you get
- One install: pi install git:github.com/memtro-org/pi-memtro, then pi --model auto:chat
- Or by hand: a provider in ~/.pi/agent/models.json with baseUrl https://platform.memtro.com/v1 and api openai-completions
- Remote MCP server, from the extension or from pi mcp add — tools arrive as mcp__memtro__*
- Live model discovery, thinking levels mapped to reasoning_effort, and memtro options through samplingParams
How to connect
- 01Create a personal API key on Dashboard → Profile → API keys and export it as MEMTRO_API_KEY
- 02pi install git:github.com/memtro-org/pi-memtro, then pi --model auto:chat
- 03Or add the memtro provider to ~/.pi/agent/models.json and run pi mcp add memtro --url https://platform.memtro.com/api/mcp --bearer-token-env-var MEMTRO_API_KEY
Example queries
- › Use memtro to find who owns the billing migration
- › Look up Help Scout ticket 2450 through memtro and check the customer in HubSpot
Why it is better with memory
Through Pi you get the same memory, connector tools, routing and jobs as every other client, so what one person learns in one tool is available to the whole team in all of them.