Cokodo Agent
Expose your repo’s `.agent/` context, references, and collaboration metadata to Claude Code, Cursor, or other MCP IDEs so the agent reasons from structured project protocol instead of guessing from lo
Overview
Cokodo Agent is a MCP server for the Build phase that serves `.agent/` project context, references, and collaborations to MCP-compatible IDEs over stdio.
What is this MCP server?
- Stdio MCP server (`cokodo-agent` on PyPI, v1.7.0) for MCP-compatible IDEs
- Serves `.agent/` project context, references, and collaborations from the repo
- Backed by the agent_protocol GitHub project (dinwind/agent_protocol)
- stdio transport—typical Claude Desktop / Claude Code MCP config pattern
- Keeps agent instructions and refs centralized instead of duplicating them in every chat
- Server version 1.7.0 on PyPI identifier `cokodo-agent`
- stdio MCP transport
- Source: github.com/dinwind/agent_protocol
What problem does it solve?
Agents in the IDE keep missing or misreading your canonical `.agent/` protocol because context is scattered across chats and manual @-mentions.
Who is it for?
Indie builders on Claude Code or Cursor who maintain a `.agent/` or agent_protocol tree and want one MCP hook instead of re-uploading context.
Skip if: Teams without an `.agent/` convention or anyone who only needs generic file search—use native workspace tools instead.
What do I get? / Deliverables
Your MCP client loads structured `.agent/` context on demand so coding agents align with the same references and collaboration metadata your repo defines.
- Live MCP tools/resources backed by `.agent/` context and references
- Reduced manual context paste into agent chats
- Consistent collaboration metadata visible to the IDE agent
Recommended MCP Servers
Journey fit
Canonical shelf is Build because the server wires local agent protocol assets into the IDE during active product development, not during launch or ops monitoring. Agent-tooling fits serving `.agent/` directories and collaboration hooks—the same layer where you configure how coding agents see project truth.
How it compares
MCP protocol bridge for `.agent/` assets, not a brainstorming or planning skill.
Common Questions / FAQ
Who is Cokodo Agent for?
Solo and small-team developers using MCP IDEs who standardize agent instructions and refs under `.agent/` and want the model to read that tree reliably.
When should I use Cokodo Agent?
Use it during active Build work whenever you rely on agent_protocol-style context and need Claude or Cursor to pull references and collaborations without manual copy-paste.
How do I add Cokodo Agent to my agent?
Install the `cokodo-agent` PyPI package (v1.7.0), add a stdio MCP server entry pointing at that CLI in Claude Code or Claude Desktop config, and open a repo that contains `.agent/`.