What is MCP
Model Context Protocol (MCP) is an open protocol from Anthropic that gives AI agents a uniform interface to any external data source or tool — filesystems, git repos, databases, APIs, design files, Slack messages…
mimo-tui v0.2 ships a complete MCP client, so you can reuse any MCP server already written for Claude Code / Codex / Cursor.
Why it matters
Without MCP, every new integration means writing agent code with bespoke schemas. With MCP:
- Install an MCP server (npm / pip / brew)
- Add one entry to
~/.mimo/mcp.json - Restart
mimo— its tools appear, the agent can call them
Current support
| Transport | Status |
|---|---|
| stdio (subprocess) | ✅ v0.2 |
| streamable HTTP | ⏳ v0.3 |
| SSE | ⏳ v0.3 |
The vast majority of official MCP servers use stdio, so stdio is enough.
Native tools vs MCP tools
mimo ships 8 native tools (read_file / shell / glob / etc.). MCP tools are additional, namespaced as mcp__<server>__<tool> to avoid clashes:
read_file ← built-inmcp__fs__read_file ← from MCP filesystem serverThe model picks. Rule of thumb:
- Built-ins: common operations (read files, run shell)
- MCP: external systems (databases, advanced git, network services)