You said no MCP
The Pi coding agent team reversed its public stance and added Model Context Protocol (MCP) support, integrating it with a new JavaScript sandbox called Codemode.
summary is enough The post explains a notable engineering reversal and introduces Codemode, but the core points fit comfortably in a brief summary.
summary
Pi.dev previously rejected the Model Context Protocol (MCP), but has now reversed position to add native support. The maintainers explain that MCP has matured and that combining it with Codemode—a JavaScript sandbox running on the harness side—solves traditional composition issues. This integration allows agents to chain multiple tools together without cluttering the context window or requiring excessive round-trips. Participants also debate the broader utility of MCP for non-coding tasks versus traditional CLI and shell scripts.
key points
- Pi integrated MCP after previously rejecting it, citing protocol maturation and shared requirements with their internal interpreter architecture.
- Codemode introduces a JavaScript sandbox running on the harness side, allowing agents to chain and coordinate tool calls locally rather than through round-trip context hops.
- The combination allows tools like Linear and Jev to be composed safely without wasting context window tokens.
who should read it
Developers building AI agent harnesses or evaluating tool protocols like MCP.
discussion positive
consensus
Most participants focus heavily on whether MCP is necessary or if standard shell commands are superior, highlighting MCP's growing value for desktop automation.
pushback
Some technical users question why a custom JavaScript Codemode sandbox is needed when modern models already excel at writing bash or python scripts to chain tools.
The value is that rather than an agent chaining together tool calls itself (which means each step sends the result back to the agent for it to analyse and work out what to do next), it writes a script for the harness to execute that chains together all the calls. The major benefits are:
* speed - much fewer hops back to the LLM
* fewer tokens - intermediate execution steps in the script don't leak into context, only the final result does.
* repeatability - if the LLM needs to repeat work, it can reuse a script it wrote last time.
If you have a harness that has access to a full shell and …
It clearly breaks down the practical advantages of Codemode over direct agent-driven tool chaining in terms of speed, token efficiency, and script reuse.
Generated by an LLM from the article and thread. Check the sources before relying on it.