# cody-mcp

A [Model Context Protocol](https://modelcontextprotocol.io) server for **[Cody](https://agentcody.ai)** — AI teammates that work inside the tools you already use.

`cody-mcp` exposes Cody's integration catalog, team-knowledge search and product answers as MCP tools, so any MCP-capable client (Claude Desktop, Claude Code, Cursor, VS Code, custom agents) can call them directly.

> Docs and product: **https://agentcody.ai**

## Tools

| Tool | Input | What it does |
| --- | --- | --- |
| `search_team_knowledge` | `query` (string), `limit` (1–10, optional) | Ranked search over Cody's knowledge base (setup, support automation, MCP, integrations). Returns titles, summaries and canonical URLs. |
| `list_integrations` | `category` (optional), `search` (optional) | Lists the tools Cody connects to — Zendesk, Jira, Slack, HubSpot, Airtable, Mixpanel and more — with categories and docs links. |
| `ask_cody` | `question` (string) | Answers product questions (pricing, setup, integrations, MCP support, data handling) with the source URLs it drew from. |

Every tool returns both human-readable text and a machine-readable `structuredContent` payload. All three are deterministic and fully offline — no API keys, no network calls, no stubs that throw.

## Install

```bash
npm install -g cody-mcp
```

Or run it without installing:

```bash
npx -y cody-mcp --version
```

Requires Node.js 18 or newer.

## Configure your MCP client

Add the server to your client's MCP configuration.

**Claude Desktop** (`claude_desktop_config.json`) / **Claude Code** (`.mcp.json`) / **Cursor** (`~/.cursor/mcp.json`):

```json
{
  "mcpServers": {
    "cody": {
      "command": "npx",
      "args": ["-y", "cody-mcp"]
    }
  }
}
```

Restart the client and the three `cody-mcp` tools become available to the model.

## Example

Ask your client:

> Use cody-mcp to find out how Cody handles Zendesk ticket triage, then list the support integrations.

The model calls `search_team_knowledge` with `{"query": "zendesk ticket triage"}` and `list_integrations` with `{"category": "Support"}`, and answers from the structured results.

## Development

```bash
git clone https://github.com/daolmedo/cody-mcp.git
cd cody-mcp
npm install
npm run build
npm run smoke     # spawns the server over stdio and calls all three tools
```

The smoke test starts the compiled server as a real MCP stdio process, performs the MCP handshake, lists the tools and invokes each one, asserting structured results come back.

## Registry manifest

`server.json` describes the package in the format MCP registries consume.

## License

MIT

---

Cody — **https://agentcody.ai**