README.md•14.7 kB
# MCP Outline Server
[](https://pypi.org/project/mcp-outline/)
[](https://www.python.org/downloads/)
[](LICENSE)
[](https://github.com/Vortiago/mcp-outline/actions/workflows/ci.yml)
[](https://github.com/Vortiago/mcp-outline/pkgs/container/mcp-outline)
A Model Context Protocol server for interacting with Outline document management.
## Features
- **Document operations**: Search, read, create, edit, archive documents
- **Collections**: List, create, manage document hierarchies
- **Comments**: Add and view threaded comments
- **Backlinks**: Find documents referencing a specific document
- **MCP Resources**: Direct content access via URIs (outline://document/{id}, outline://collection/{id}, etc.)
- **Automatic rate limiting**: Transparent handling of API limits with retry logic
## Prerequisites
Before using this MCP server, you need:
- An [Outline](https://www.getoutline.com/) account (cloud hosted or self-hosted)
- API key from Outline web UI: **Settings → API Keys → Create New**
- Python 3.10+ (for non-Docker installations)
> **Getting your API key**: Log into Outline → Click your profile → Settings → API Keys → "New API Key". Copy the generated token.
## Installation
### Using uv (Recommended)
```bash
uvx mcp-outline
```
### Using pip
```bash
pip install mcp-outline
```
### Using Docker
```bash
docker run -e OUTLINE_API_KEY=<your-key> ghcr.io/vortiago/mcp-outline:latest
```
Or build from source:
```bash
docker buildx build -t mcp-outline .
docker run -e OUTLINE_API_KEY=<your-key> mcp-outline
```
## Configuration
| Variable | Required | Default | Notes |
|----------|----------|---------|-------|
| `OUTLINE_API_KEY` | Yes | - | Get from Outline web UI: Settings → API Keys → Create New |
| `OUTLINE_API_URL` | No | `https://app.getoutline.com/api` | For self-hosted: `https://your-domain/api` |
| `OUTLINE_READ_ONLY` | No | `false` | `true` = disable ALL write operations ([details](#read-only-mode)) |
| `OUTLINE_DISABLE_DELETE` | No | `false` | `true` = disable only delete operations ([details](#disable-delete-operations)) |
| `OUTLINE_DISABLE_AI_TOOLS` | No | `false` | `true` = disable AI tools (for Outline instances without OpenAI) |
| `MCP_TRANSPORT` | No | `stdio` | Transport mode: `stdio` (local), `sse` or `streamable-http` (remote) |
| `MCP_HOST` | No | `127.0.0.1` | Server host. Use `0.0.0.0` in Docker for external connections |
| `MCP_PORT` | No | `3000` | HTTP server port (only for `sse` and `streamable-http` modes) |
## Access Control
Configure server permissions to control what operations are allowed:
### Read-Only Mode
Set `OUTLINE_READ_ONLY=true` to enable viewer-only access. Only search, read, export, and collaboration viewing tools are available. All write operations (create, update, move, archive, delete) are disabled.
**Use cases:**
- Shared access for team members who should only view content
- Safe integration with AI assistants that should not modify documents
- Public or demo instances where content should be protected
**Available tools:**
- Search & Discovery: `search_documents`, `list_collections`, `get_collection_structure`, `get_document_id_from_title`
- Document Reading: `read_document`, `export_document`
- Comments: `list_document_comments`, `get_comment`
- Collaboration: `get_document_backlinks`
- Collections: `export_collection`, `export_all_collections`
- AI: `ask_ai_about_documents` (if not disabled with `OUTLINE_DISABLE_AI_TOOLS`)
### Disable Delete Operations
Set `OUTLINE_DISABLE_DELETE=true` to allow create and update workflows while preventing accidental data loss. Only delete operations are disabled.
**Use cases:**
- Production environments where documents should not be deleted
- Protecting against accidental deletions
- Safe content editing workflows
**Disabled tools:**
- `delete_document`, `delete_collection`
- `batch_delete_documents`
**Important:** `OUTLINE_READ_ONLY=true` takes precedence over `OUTLINE_DISABLE_DELETE`. If both are set, the server operates in read-only mode.
## Adding to Your Client
> **Prerequisites**: Install `uv` with `pip install uv` or from [astral.sh/uv](https://docs.astral.sh/uv/)
<details>
<summary><b>Add to Claude Desktop</b></summary>
Edit `~/Library/Application Support/Claude/claude_desktop_config.json` (or `%APPDATA%\Claude\claude_desktop_config.json` on Windows):
```json
{
"mcpServers": {
"mcp-outline": {
"command": "uvx",
"args": ["mcp-outline"],
"env": {
"OUTLINE_API_KEY": "<YOUR_API_KEY>",
"OUTLINE_API_URL": "<YOUR_OUTLINE_URL>" // Optional
}
}
}
}
```
</details>
<details>
<summary><b>Add to Cursor</b></summary>
Go to **Settings → MCP** and click **Add Server**:
```json
{
"mcp-outline": {
"command": "uvx",
"args": ["mcp-outline"],
"env": {
"OUTLINE_API_KEY": "<YOUR_API_KEY>",
"OUTLINE_API_URL": "<YOUR_OUTLINE_URL>" // Optional
}
}
}
```
</details>
<details>
<summary><b>Add to VS Code</b></summary>
Create a `.vscode/mcp.json` file in your workspace with the following configuration:
```json
{
"servers": {
"mcp-outline": {
"type": "stdio",
"command": "uvx",
"args": ["mcp-outline"],
"env": {
"OUTLINE_API_KEY": "<YOUR_API_KEY>"
}
}
}
}
```
For self-hosted Outline instances, add `OUTLINE_API_URL` to the `env` object.
**Optional**: Use input variables for sensitive credentials:
```json
{
"inputs": [
{
"type": "promptString",
"id": "outline-api-key",
"description": "Outline API Key",
"password": true
}
],
"servers": {
"mcp-outline": {
"type": "stdio",
"command": "uvx",
"args": ["mcp-outline"],
"env": {
"OUTLINE_API_KEY": "${input:outline-api-key}"
}
}
}
}
```
VS Code will automatically discover and load MCP servers from this configuration file. For more details, see the [official VS Code MCP documentation](https://code.visualstudio.com/docs/copilot/chat/mcp-servers).
</details>
<details>
<summary><b>Add to Cline (VS Code)</b></summary>
In Cline extension settings, add to MCP servers:
```json
{
"mcp-outline": {
"command": "uvx",
"args": ["mcp-outline"],
"env": {
"OUTLINE_API_KEY": "<YOUR_API_KEY>",
"OUTLINE_API_URL": "<YOUR_OUTLINE_URL>" // Optional
}
}
}
```
</details>
<details>
<summary><b>Using pip instead of uvx</b></summary>
If you prefer to use `pip` instead:
```bash
pip install mcp-outline
```
Then in your client config, replace `"command": "uvx"` with `"command": "mcp-outline"` and remove the `"args"` line:
```json
{
"mcp-outline": {
"command": "mcp-outline",
"env": {
"OUTLINE_API_KEY": "<YOUR_API_KEY>",
"OUTLINE_API_URL": "<YOUR_OUTLINE_URL>" // Optional
}
}
}
```
</details>
<details>
<summary><b>Docker Deployment (HTTP)</b></summary>
For remote access or Docker containers, use HTTP transport. This runs the **MCP server** on port 3000:
```bash
docker run -p 3000:3000 \
-e OUTLINE_API_KEY=<YOUR_API_KEY> \
-e MCP_TRANSPORT=streamable-http \
ghcr.io/vortiago/mcp-outline:latest
```
Then connect from client:
```json
{
"mcp-outline": {
"url": "http://localhost:3000/mcp"
}
}
```
**Note**: `OUTLINE_API_URL` should point to where your Outline instance is running, not localhost:3000.
</details>
## Tools
> **Note**: Tool availability depends on your [Access Control](#access-control) settings. Some tools are disabled in read-only mode or when delete operations are restricted.
### Search & Discovery
- `search_documents(query, collection_id?, limit?, offset?)` - Search documents by keywords with pagination
- `list_collections()` - List all collections
- `get_collection_structure(collection_id)` - Get document hierarchy within a collection
- `get_document_id_from_title(query, collection_id?)` - Find document ID by title search
### Document Reading
- `read_document(document_id)` - Get document content
- `export_document(document_id)` - Export document as markdown
### Document Management
- `create_document(title, collection_id, text?, parent_document_id?, publish?)` - Create new document
- `update_document(document_id, title?, text?, append?)` - Update document (append mode available)
- `move_document(document_id, collection_id?, parent_document_id?)` - Move document to different collection or parent
### Document Lifecycle
- `archive_document(document_id)` - Archive document
- `unarchive_document(document_id)` - Restore document from archive
- `delete_document(document_id, permanent?)` - Delete document (or move to trash)
- `restore_document(document_id)` - Restore document from trash
- `list_archived_documents()` - List all archived documents
- `list_trash()` - List all documents in trash
### Comments & Collaboration
- `add_comment(document_id, text, parent_comment_id?)` - Add comment to document (supports threaded replies)
- `list_document_comments(document_id, include_anchor_text?, limit?, offset?)` - View document comments with pagination
- `get_comment(comment_id, include_anchor_text?)` - Get specific comment details
- `get_document_backlinks(document_id)` - Find documents that link to this document
### Collection Management
- `create_collection(name, description?, color?)` - Create new collection
- `update_collection(collection_id, name?, description?, color?)` - Update collection properties
- `delete_collection(collection_id)` - Delete collection
- `export_collection(collection_id, format?)` - Export collection (default: outline-markdown)
- `export_all_collections(format?)` - Export all collections
### Batch Operations
- `batch_create_documents(documents)` - Create multiple documents at once
- `batch_update_documents(updates)` - Update multiple documents at once
- `batch_move_documents(document_ids, collection_id?, parent_document_id?)` - Move multiple documents
- `batch_archive_documents(document_ids)` - Archive multiple documents
- `batch_delete_documents(document_ids, permanent?)` - Delete multiple documents
### AI-Powered
- `ask_ai_about_documents(question, collection_id?, document_id?)` - Ask natural language questions about your documents
## Resources
- `outline://collection/{id}` - Collection metadata (name, description, color, document count)
- `outline://collection/{id}/tree` - Hierarchical document tree structure
- `outline://collection/{id}/documents` - Flat list of documents in collection
- `outline://document/{id}` - Full document content (markdown)
- `outline://document/{id}/backlinks` - Documents that link to this document
## Development
### Quick Start with Self-Hosted Outline
```bash
# Generate configuration
cp config/outline.env.example config/outline.env
openssl rand -hex 32 > /tmp/secret_key && openssl rand -hex 32 > /tmp/utils_secret
# Update config/outline.env with generated secrets
# Start all services
docker compose up -d
# Create API key: http://localhost:3030 → Settings → API Keys
# Add to .env: OUTLINE_API_KEY=<token>
```
### Setup
```bash
git clone https://github.com/Vortiago/mcp-outline.git
cd mcp-outline
uv pip install -e ".[dev]"
```
### Testing
```bash
# Run tests
uv run pytest tests/
# Format code
uv run ruff format .
# Type check
uv run pyright src/
# Lint
uv run ruff check .
```
### Running Locally
```bash
uv run mcp-outline
```
### Testing with MCP Inspector
Use the MCP Inspector to test the server tools visually via an interactive UI.
**For local development** (with stdio):
```bash
npx @modelcontextprotocol/inspector -e OUTLINE_API_KEY=<your-key> -e OUTLINE_API_URL=<your-url> uv run python -m mcp_outline
```
**For Docker Compose** (with HTTP):
```bash
npx @modelcontextprotocol/inspector http://localhost:3000
```

## Architecture Notes
**Rate Limiting**: Automatically handled via header tracking (`RateLimit-Remaining`, `RateLimit-Reset`) with exponential backoff retry (up to 3 attempts). No configuration needed.
**Transport Modes**:
- `stdio` (default): Direct process communication
- `sse`: HTTP Server-Sent Events (use for web clients)
- `streamable-http`: Streamable HTTP transport
**Connection Pooling**: Shared httpx connection pool across instances (configurable: `OUTLINE_MAX_CONNECTIONS=100`, `OUTLINE_MAX_KEEPALIVE=20`)
## Troubleshooting
### Server not connecting?
**Check your API credentials:**
```bash
# Test your API key
curl -H "Authorization: Bearer YOUR_API_KEY" YOUR_OUTLINE_URL/api/auth.info
```
Common issues:
- Verify `OUTLINE_API_KEY` is set correctly in your MCP client configuration
- Check `OUTLINE_API_URL` points to your Outline instance (default: `https://app.getoutline.com/api`)
- For self-hosted Outline, ensure the URL ends with `/api`
- Verify your API key hasn't expired or been revoked
### Tools not appearing in client?
- **Read-only mode enabled?** Check if `OUTLINE_READ_ONLY=true` is disabling write tools
- **Delete operations disabled?** Check if `OUTLINE_DISABLE_DELETE=true` is hiding delete tools
- **AI tools missing?** Check if `OUTLINE_DISABLE_AI_TOOLS=true` is disabling AI features
- Restart your MCP client after changing environment variables
### API rate limiting errors?
The server automatically handles rate limiting with retry logic. If you see persistent rate limit errors:
- Reduce concurrent operations
- Check if multiple clients are using the same API key
- Contact Outline support if limits are too restrictive for your use case
### Docker container issues?
**Container won't start:**
- Ensure `OUTLINE_API_KEY` is set: `docker run -e OUTLINE_API_KEY=your_key ...`
- Check logs: `docker logs <container-id>`
**Can't connect from client:**
- Use `0.0.0.0` for MCP_HOST: `-e MCP_HOST=0.0.0.0`
- Verify port mapping: `-p 3000:3000`
- Check transport mode: `-e MCP_TRANSPORT=streamable-http`
### Need more help?
- 📖 [MCP Documentation](https://modelcontextprotocol.io/)
- 🐛 [Report an issue](https://github.com/Vortiago/mcp-outline/issues)
## Contributing
Contributions welcome! Please submit a Pull Request.
## License
This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.
## Acknowledgments
- Built with [MCP Python SDK](https://github.com/modelcontextprotocol/python-sdk)
- Uses [Outline API](https://getoutline.com) for document management