> ## Documentation Index
> Fetch the complete documentation index at: https://judged.systems/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP

> Connect an MCP client with OAuth and judge a published pack.

The MCP server is `{origin}/mcp`. A client signs in through OAuth. An API key is not accepted.

```json theme={null}
{
  "mcpServers": {
    "judged": {
      "url": "https://judged.systems/mcp"
    }
  }
}
```

The first connection opens `/login`, then `/consent`. Approve to issue a token for your user. The token can run the same actions as a signed-in session: read and write packs, run and read judgments, run simulations, and write labels. The four tools below are the only ones exposed.

## Tools

| Tool | Input | Returns |
| - | - | - |
| `judge_ticket` | `packId`, `ticket` | The same `{ evaluationId, result, packVersionId }` as `POST /api/judge` |
| `get_evaluation` | `evaluationId` | The stored evaluation, including the redacted ticket and labels |
| `list_published_packs` | none | Packs that have a live version |
| `get_pack` | `packId` | The live published document |

`judge_ticket` and `get_pack` use the published version. A pack with no published version fails the tool call. A pack or evaluation that belongs to another user is not returned.

`ticket` is the [ticket](/docs/guides/concepts) object. One call judges one ticket.

## What the tools do not do

They do not accept a draft. They do not write labels. They do not start a simulation. Use REST for those, with an API key.

Playground remains the place to judge a draft before you publish.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.