# agloop-mcp

MCP server providing 14 agent introspection tools for [AgLoop](https://github.com/agloop/agloop) — the VS Code Copilot custom agent framework.

## Quick Start

Add to your `.vscode/mcp.json`:
> For claude desktop or cli users use `mcpservers` instead of `servers`

```json
{
  "servers": {
    "agloop": {
      "command": "uvx",
      "args": ["agloop-mcp"]
    }
  }
}
```

Or run directly:

```bash
uvx agloop-mcp
```

## Tools

### Read Tools (10)

| Tool                            | Description                                                    |
| ------------------------------- | -------------------------------------------------------------- |
| `agloop_get_state`              | Full AgLoop state: phase, tasks, iteration, compaction context |
| `agloop_get_task`               | Single task details by ID                                      |
| `agloop_list_tasks`             | All tasks with optional status/dependency filter               |
| `agloop_get_logs`               | Execution log entries with agent/task/limit filters            |
| `agloop_get_plan`               | Current plan (YAML or Markdown)                                |
| `agloop_get_compaction_context` | Compaction recovery context                                    |
| `agloop_search_logs`            | Full-text search through execution logs                        |
| `agloop_get_agent_info`         | Agent definition file (.agent.md) content                      |
| `agloop_get_next_task`          | Topological-sort next task recommendation                      |
| `agloop_list_agents`            | All available agent names                                      |

### Write Tools (4)

| Tool                       | Description                              |
| -------------------------- | ---------------------------------------- |
| `agloop_update_task`       | Update task status and result log        |
| `agloop_append_log`        | Append structured entry to execution log |
| `agloop_create_checkpoint` | Snapshot current state for recovery      |
| `agloop_set_phase`         | Transition to a new phase                |

## Configuration

The server uses the current working directory as the workspace root. Override with:

```bash
uvx agloop-mcp --workspace /path/to/project
```

## State Files

The server reads/writes from `.agloop/` in the workspace root:

```
.agloop/
├── state.json        # Phase, tasks, iteration
├── log.json          # Execution log (JSON Lines)
├── plan.yaml         # Current plan
└── checkpoints/      # State snapshots
```

## Requirements

- Python 3.11+
- No additional dependencies beyond `mcp`

## License

MIT