Pi coding agent reverses course and adds MCP support with 1.0 release

The Pi coding agent reached its 1.0 milestone on Thursday, and the headline addition is support for the Model Context Protocol (MCP) — a notable reversal, since Pi's creator Mario Zechner once dismissed MCP as unnecessary.

Earendil, the company formed by Armin Ronacher and Colin Daymond Hanna that acquired Pi in April 2026, explained the about-face by noting that things change. Not only has MCP itself improved, the company said, but the changes Pi's developers made to MCP have made integrating other capabilities easier. “For example, the changes we have made to MCP also enable the use of Jev more easily within Pi,” Earendil said, referring to the widely noted Jev decision model.

Earendil describes Pi as “a hardened, minimal, extensible agent harness that you can make your own.” The Register's analogy is a Python web framework: if Pi were Flask, it would not be Django — light and malleable rather than all-encompassing.

The 1.0 release still packs in new features. These include Codemode, a harness-side sandbox that helps agents issue tool calls, with support for MCP, decision models, and image models, plus extension support for virtual models, deferred tool loading, cache warming for Anthropic models, mid-conversation system messages, a new TUI theme, and full-screen mode by default.

Alongside those features, Earendil built Pi Durable, described as “a new substrate for building long-running agentic applications.” It is a distinct component from Pi, the company explained, in order to preserve Pi's minimalist design.

The division of labor is deliberate: the Pi agent connects the underlying model to tools to manage conversation, while Pi Durable is the layer that orchestrates multiple conversations and handles the storage backend via SQLite and JSONL. “It does not replace the Pi coding agent,” Earendil said. “It is a framework for building any agentic application, coding agents included,” sharing both code such as pi-ai and the principles of minimalism and malleability.

The reversal matters because it reflects how MCP's standing in the agent ecosystem has shifted. A protocol once treated as optional — even dismissed as redundant — is now being adopted by default across more tools, and a harness that sells itself on minimalism has chosen to add a dedicated layer for it.

What to watch next is whether Pi's MCP implementation and the Codemode sandbox can lower integration costs without bloating the tool, and whether Pi Durable draws developers building long-running agents beyond coding.