Agents

OpenCodeHub gives an AI coding agent a code graph it can query: callers, callees, processes, blast radius, owners, scanner findings, and a 5-tier PR verdict — all behind 29 MCP tools served by one local binary. The graph is built deterministically from your repo and stored next to it.

Other docs sections answer “what is OCH” and “how is it built.” This section answers “how does my agent talk to it.”

Who this is for

• An AI coding agent (Claude Code, Cursor, Codex, Windsurf, OpenCode, or anything else that speaks MCP) that needs to ground its answers in the structure of a codebase.
• The engineer wiring that agent up.

If you are reading this with an LLM in the loop, the rest of this section is paste-ready. Headings are scannable. Code is first-class.

90-second setup

install OpenCodeHub once, per-machine

git clone https://github.com/theagenticguy/opencodehub
cd opencodehub
pnpm install --frozen-lockfile
mise run cli:link        # puts `codehub` on PATH

wire it into a target repo

cd /path/to/your/repo
codehub init             # writes .mcp.json + links the Claude Code plugin
codehub analyze          # first index — 30s to a few minutes

Restart your editor. Your agent now has 29 MCP tools, all prefixed mcp__opencodehub__*. See Install for the full path or jump to the per-editor card below.

Pick your editor

Claude Code Plugin + 5 slash commands + code-analyst subagent + 11 skills.

Cursor Per-project .cursor/mcp.json — MCP only, no plugin.

Codex codex.toml MCP entry — paste once, works in CLI and IDE.

Windsurf mcp_config.json — Cascade can call all 29 tools after restart.

OpenCode opencode.json — local stdio, zero auth.

What’s in this section

Why MCP What an agent cannot see without a code graph.

Install Generic install path that works for any MCP-speaking agent.

Tool decision matrix Intent in, tool out. Anti-patterns called out.

Idiomatic prompts Five copy-paste prompts that get great use out of OCH.

Discovery and resources Where agents and operators find OCH on the open web.

MCP registries One-click install paths via Smithery and friends.

llms.txt cheatsheet Which crawlable bundle to feed an agent in which scenario.