# Overview

> Connect AI agents to your Clad workspace with the hosted Model Context Protocol server — same permissions and object ids as the API.

Product: Clad MCP
Source: https://docs.useclad.ai/mcp/overview

---

The MCP server lets any MCP-compatible agent act in your Clad workspace with the
same permission model as the API.

```
https://clad-server-production.up.railway.app/mcp
```

- **Transport**: Streamable HTTP — the current MCP standard. There is no local
  install and no stdio bridge; point your client at the URL above.
- **Tools, not schema**: the server exposes more than 60 purpose-built tools
  (`search_issues`, `reply_to_issue`, `create_article`, …) with typed inputs and
  human-readable descriptions. Agents call these far more reliably than they
  navigate a raw GraphQL schema.
- **Same objects, same ids**: every id an MCP tool returns is a global object id
  that works verbatim in the GraphQL API, and vice versa — so you can start in an
  agent and finish in a script without translating ids.

Two ways to connect, depending on who is acting:

| | Who acts | Auth | Best for |
|---|---|---|---|
| **OAuth** | A person, through their agent | Sign in with your Clad account | Cursor, Claude, ChatGPT, day-to-day agent use |
| **API key** | A headless service | `Authorization: Bearer clad_mk_...` | Backend automation, cron jobs, CI |

Copy-ready connection snippets for every client live in **Settings → MCP** in the
Clad app.