Earendil reversed its stance and added MCP support to Pi. MCP itself evolved, and Pi’s own needs aligned with MCP’s requirements. MCP in Pi now treats tools like OpenAPI, with structured data and discoverability. Composition remains MCP’s main weakness, but the issue lies with servers and harnesses. Pi’s Codemode provides a JavaScript sandbox for orchestrating tool calls. Codemode runs on the trusted harness side and maintains state in the session. Codemode is loaded automatically when MCP is configured in Pi. Users can enable Codemode manually via Pi’s configuration. Codemode allows combining MCP tools like Linear and Jev for complex tasks. Jev can classify issue tracker comments by emotional tone within Pi. Earendil aims to shape MCP’s future by embracing and improving it.
What changed
Pi explicitly did not support MCP and had dismissive public statements about it.
What it unlocks
Using MCP servers directly in Pi with structured tool discovery and Codemode orchestration.
What you need to act on it
- Pi installed and updated
- MCP server configured
Sources