Pi coding agent pulls a 180 and adds MCP support
Pi, a minimalist coding agent, reached version 1.0 on Thursday with a significant technical reversal: the Earendil-owned project now supports Model Context Protocol (MCP), a technology that creator Mario Zechner once dismissed as unnecessary. The company explained the about-face by noting that MCP has improved substantially and that changes the Pi team implemented in the protocol have made integrating additional capabilities considerably easier. The move also enables better integration with tools like Jev, a decision model that has already inspired imitators.
Alongside MCP support, Pi 1.0 includes several new features: Codemode (a sandbox for tool calls), extension support for virtual models, deferred tool loading, cache warming for Anthropic models, mid-conversation system messages, and full-screen mode by default. Earendil has also introduced Pi Durable, a separate component designed for long-running agentic applications that orchestrates multiple conversations and handles storage via SQLite and JSONL, whilst preserving Pi's minimalist design philosophy.
- Pi coding agent reaches 1.0 with MCP support, reversing founder's earlier position
- New features include Codemode sandbox and virtual model extensions
- Pi Durable launched to handle long-running applications whilst maintaining minimalism
New here? Start with this
Pi is a simple software tool that helps developers write computer code. Its creator, Mario Zechner, had previously dismissed Model Context Protocol (MCP) as unnecessary. Pi's recent version 1.0 release now includes support for MCP.
MCP is a technical standard that allows different software tools to communicate and work together effectively. The shift to support it came because MCP has improved substantially since Zechner's earlier assessment, and the Pi team made changes to MCP that made integration simpler and more aligned with Pi's minimalist philosophy.
Earendil, the company that owns Pi, has long valued minimalism and simplicity in the tool's design. As MCP has evolved and the Pi team has refined it, the technology has become compatible with Pi's principles whilst enabling integration with tools like Jev.