Jules MCP Server
# Jules MCP Server
[](https://opensource.org/licenses/MIT)
[](https://www.python.org/downloads/)
[](https://github.com/jlowin/fastmcp)
[](https://github.com/cavuminfundo/jules-mcp-server/pkgs/container/jules-mcp-server)
Production-grade [Model Context Protocol (MCP)](https://modelcontextprotocol.io) server for [Google Jules AI](https://jules.google.com), built with [FastMCP](https://github.com/jlowin/fastmcp).
Provides an enterprise-ready bridge for AI Agents (Antigravity, Claude Desktop, Cursor, VS Code, Goose) to inspect, mentor, plan, and automate Google Jules execution sessions natively via HTTP, SSE, or Stdio transport.
---
## Key Features
- โก **Sub-Second Response Times**: Single-page fast listing (`fetch_all=False` by default) returning session data in <200ms.
- ๐ก๏ธ **Hardened Network Resilience**: Per-request isolated HTTP clients (`httpx.AsyncClient`) with strict timeouts (`httpx.Timeout(8.0)`) catching `asyncio.CancelledError` and `TimeoutError` to guarantee non-blocking agent turns.
- ๐งน **Concurrent Automatic Session Cleanup**: Parallel scanning and deletion of terminal/inactive sessions (`COMPLETED`, `FINISHED`, `TERMINATED`, `CANCELLED`, `FAILED`, `EXPIRED`, `CLOSED`) using `asyncio.gather` (`clean_completed_sessions`).
- ๐ **Type Sanitization**: Primitive string and boolean defaults (`title: ""`, `starting_branch: "main"`, `fetch_all: false`) eliminating `null` validation errors in client schema parsers.
- ๐ **Multi-Transport Support**: Native HTTP JSON-RPC (`/mcp`), Server-Sent Events (`/sse`), and Stdio bridge support.
- ๐ณ **Production Docker Image**: Pre-compiled multi-arch image hosted on GitHub Container Registry (`ghcr.io/cavuminfundo/jules-mcp-server:latest`).
---
## ๐ ๏ธ MCP Tools Reference
| Tool Name | Description | Default Parameters |
| --- | --- | --- |
| `list_sessions` | List Jules sessions with fast single-page default or optional auto-pagination. | `page_size: 50`, `page_token: ""`, `fetch_all: false` |
| `get_session` | Retrieve full details and current state for a specific session ID or name. | `session_id: string` |
| `create_session` | Launch a new Jules AI session for a target repository. | `source: string`, `prompt: string`, `title: ""`, `starting_branch: "main"`, `require_plan_approval: false` |
| `list_activities` | Fetch activity history and generated plans for a session. | `session_id: string`, `page_size: 20`, `page_token: ""` |
| `list_all_activities` | Fetch all activity history for a session using automatic pagination. | `session_id: string` |
| `get_activity` | Retrieve details for a specific activity ID within a session. | `session_id: string`, `activity_id: string` |
| `approve_session_plan` | Approve a pending session plan in a single native call. | `session_id: string` |
| `send_session_message` | Send mentoring feedback, answers, or directives to an active session. | `session_id: string`, `message: string` |
| `delete_session` | Permanently delete a completed or inactive session. | `session_id: string` |
| `clean_completed_sessions` | Concurrently scan and delete all terminal/inactive sessions. | *None* |
| `list_sources` | List accessible source GitHub repositories connected to Jules. | `page_size: 50`, `page_token: ""`, `filter_str: ""` |
| `get_all_sources` | Retrieve all accessible source repositories with auto-pagination. | `filter_str: ""` |
---
## ๐ Quickstart & Container Deployment
### 1. Docker Compose (Recommended Production Setup)
Create a `docker-compose.yml` file:
```yaml
services:
jules-mcp:
image: ghcr.io/cavuminfundo/jules-mcp-server:latest
container_name: jules_mcp_server
dns:
- 8.8.8.8
- 8.8.4.4
environment:
- JULES_API_KEY=your_google_jules_api_key_here
ports:
- "8000:8000"
restart: unless-stopped
```
Run the container:
```bash
docker compose up -d
```
### 2. Standalone Docker Run
```bash
docker run -d \
--name jules_mcp_server \
-e JULES_API_KEY="your_google_jules_api_key_here" \
-p 8000:8000 \
--restart unless-stopped \
ghcr.io/cavuminfundo/jules-mcp-server:latest
```
---
## ๐ Client Integration Guide (`mcp_config.json`)
Add `jules-mcp` to your AI client configuration (Antigravity, Claude Desktop, Cursor, VS Code, Goose).
### Option A: Stateless HTTP JSON-RPC Transport (Recommended - Zero Session State)
Stateless endpoint (`/rpc`) executing tool calls instantly without stateful SSE connection ID caching or reconnection timeouts:
```json
{
"mcpServers": {
"jules-mcp": {
"type": "http",
"url": "http://<SERVER_IP_OR_HOST>:8000/rpc"
}
}
}
```
### Option B: Native Stdio Transport Bridge (SSH / Docker)
Direct process stdio stream execution bypassing HTTP network sockets completely:
```json
{
"mcpServers": {
"jules-mcp": {
"command": "ssh",
"args": [
"-o",
"StrictHostKeyChecking=no",
"user@<SERVER_IP_OR_HOST>",
"docker",
"exec",
"-i",
"jules_mcp_server",
"python3",
"-m",
"jules_mcp.jules_mcp",
"stdio"
]
}
}
}
```
### Option C: SSE Transport (Server-Sent Events)
Stateful stream connection for streaming clients:
```json
{
"mcpServers": {
"jules-mcp": {
"type": "sse",
"url": "http://<SERVER_IP_OR_HOST>:8000/sse"
}
}
}
```
### Stdio Transport Bridge (`mcp-remote`)
```json
{
"mcpServers": {
"jules-mcp": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"http://<SERVER_IP_OR_HOST>:8000/sse",
"--allow-http"
]
}
}
}
```
---
## ๐ ๏ธ Local Development & Testing
```bash
# Clone the repo
git clone https://github.com/cavuminfundo/jules-mcp-server.git
cd jules-mcp-server
# Set environment API key
export JULES_API_KEY="your_google_jules_api_key"
# Install dependencies and start server with uv
uv sync
uv run python -m jules_mcp.jules_mcp
```
---
## ๐ License
Distributed under the [MIT License](LICENSE). See `LICENSE` for more information.
TDQS
Scored across 13 tools
Several tools have unclear boundaries, especially list_activities vs list_all_activities and list_sources vs get_all_sources, which appear to perform nearly the same listing operation with only subtle pagination differences. Descriptions do not clearly explain when to choose one over the other.
Most names follow a consistent snake_case verb_noun pattern (list_sessions, get_session, create_session, delete_session). Minor deviations exist: get_all_sources uses 'get' for a list operation, and list_all_activities/list_activities plus list_sources/get_all_sources introduce inconsistent list vs get_all phrasing.
13 tools is reasonable for a session/activity/source management server. However, the redundant auto-pagination variants slightly inflate the count, meaning not every tool clearly earns its place.
Core session lifecycle is covered with create, get, list, delete, bulk cleanup, plan approval, and messaging. Activities and sources are read/list-oriented, which may be appropriate if they are externally managed, but there is no update or termination operation beyond delete.