integrated-harness-kit-mcp 0.3.1

MCP server for managing the tot0rokr/agents harness (Claude Code / Codex CLI / OpenCode / Gemini CLI shared config)

integrated-harness-kit-mcp

An MCP server that exposes diagnostic and (in later versions) management tools for the tot0rokr/agents harness — the shared configuration layer for Claude Code, Codex CLI, OpenCode, and Gemini CLI.

Install

uvx integrated-harness-kit-mcp

That command runs the server on stdio; pair it with a CLI agent's MCP client (see the repo root's INSTALLATION.md for ready-made config snippets per CLI).

What's available

v0.2 — read-only

Tool	Purpose
harness_status	repo presence + state of the 5 home-dir symlinks
doctor	run the harness's scripts/doctor.sh and surface its output
list_skills	enumerate registered SKILL.md entries
list_mcp_servers	enumerate canonical MCP servers in shared/mcp/servers.json
list_commands	enumerate slash commands in shared/commands/
list_subagents	enumerate sub-agents in shared/subagents/

Planned

• v0.3 — clone, install, render (mutate the harness from inside a chat)
• v0.4 — add_skill, add_mcp_server, add_command, add_subagent (+ remove_*)
• v0.5 — update, audit_drift, commit, edit_memory, edit_instruction

Repo discovery

Every tool resolves the agents repo in this order:

1. The function's repo_path argument.
2. The AGENTS_REPO environment variable.
3. ~/agents.

A candidate is accepted only if it contains scripts/install.py.

Development

cd mcp/
python3 -m venv .venv
.venv/bin/pip install -e '.'
.venv/bin/python -m unittest discover -s tests -v

Tools live in src/integrated_harness_kit_mcp/tools/ as plain functions, so the unit tests don't import mcp and don't depend on FastMCP behaviour.

License

MIT. See repo root.