# MCP Server

`ox serve` turns your workspace into a tool for AI agents. It starts a Model Context Protocol (MCP) server on stdio, exposing knowledge-base search and graph tools to Claude Code, Cursor, Windsurf, and any MCP-compatible agent.

Both the Go CLI and ox-rs include an MCP server. The `.mcp.json` snippet below works with either — they expose the same 6 tools.

## Connect

Add ox to your `.mcp.json` to auto-connect:

```json
{
  "mcpServers": {
    "ox": {
      "type": "stdio",
      "command": "ox",
      "args": ["serve"]
    }
  }
}
```

## Tools exposed

| Tool | Purpose |
|------|---------|
| `search_vault` | Hybrid BM25 + vector search across notes and repo docs |
| `get_note` | Read a note's full text by path |
| `find_related` | Graph neighbors + semantic similarity from a note |
| `get_backlinks` | Inbound wikilinks to a note |
| `vault_stats` | Index statistics (note count, edge count, tags) |
| `find_path` | Shortest graph path between two notes (requires optional Neo4j backend) |

## In practice

```
# Agent query (via Claude Code):
> Search vault for "deployment checklist"

search_vault("deployment checklist")
-> 3 results in 0.04s
-> o/what/orgs/COM/iwl/deploy-checklist.md (0.96)
-> o/when/days/2026.March.20.Thursday.md (0.84)
-> x/code/.../infra/docs/DEPLOY.md (0.79)
```

## Agent-navigable by construction

The agent traverses the same tree humans do — no special API, no wrapper. Two things make the workspace agent-ready out of the box:

- **`.agents/AGENTS.md`** provides universal instructions for any AI agent that opens the workspace.
- **The MCP server** gives agents fast, ranked retrieval over the indexed graph.

Agents are first-class citizens in OX: they receive the same context humans get.

## Next steps

- [CLI Reference](https://iamox.ai/docs/cli.md#ox-serve)
- [Two Implementations](https://iamox.ai/docs/two-implementations.md)
- [Knowledge Graph & RAG](https://iamox.ai/docs/knowledge-graph-rag.md)