LinkedIn MCP Server

# LinkedIn MCP Server with Anthropic's Claude Integration A Python-based MCP (Model Context Protocol) server that gets stuff from your LinkedIn profile and integrates with the Anthropic API for potential analysis tasks. # TL;DR Install for Claude Desktop/Code Access to the LinkedIn profile ```bash # 1.a) Install the mcp server access in Claude Desktop ./install_claude_desktop_mcp.sh # 1.b) or manually integrate this JSON snippet to the `mcpServers` section of your `claude_desktop_config.json` (e.g. `~/Library/Application\ Support/Claude/claude_desktop_config.json`) { "linkedin_francisco_perez_sorrosal": { "command": "npx", "args": ["mcp-remote", "http://localhost:10000/mcp"] } } # 2) Restart Claude and check that the 'Add from linkedin_francisco_perez_sorrosal` option is available in the mcp servers list # 3) Query the LinkedIn profile served from the mcp server in Claude Desktop! e.g. TODO ``` ## Features - **Job Search URL Generation**: Create properly formatted LinkedIn job search URLs with location, distance, and query parameters - **Job ID Retrieval**: Extract job IDs from LinkedIn search pages with pagination support - **Job Metadata Extraction**: Get detailed job information including title, company, description, and requirements - **CV Adaptation**: Combined with the cv MCP Server, adapt Francisco's CV to match specific job requirements - Built with FastMCP for high performance and with Pixi for dependency management and task running - Source code organized in the `src/` directory - Includes configurations for: - Docker (optional, for containerization) - Linting (Ruff, Black, iSort) - Formatting - Type checking (MyPy) ## Prerequisites - Python 3.13+ - [Pixi](https://pixi.sh/) (for dependency management and task execution) - uv (for building the MCP bundle) - Docker (optional, for containerization) ## Project Structure This project follows the `src` layout for Python packaging. ```bash . ├── .dockerignore ├── .gitignore ├── Dockerfile ├── pyproject.toml # Python project metadata and dependencies (PEP 621) ├── README.md ├── src/ │ └── linkedin_mcp_server/ │ ├── __init__.py │ └── main.py # FastAPI application logic ├── tests/ # Test files (e.g., tests_main.py) ``` ## Setup and Installation 1. **Clone the repository** (if applicable) or ensure you are in the project root directory. 2. **Install dependencies using Pixi**: This command will create a virtual environment and install all necessary dependencies: ```bash pixi install ``` ## Running the Server Pixi tasks are defined in `pyproject.toml`: ### mcps (MCP Server) ```bash pixi run mcps --transport stdio ``` ### Development Mode (with auto-reload) ```bash # Using pixi directly pixi run mcps --transport stdio # or sse, streamable-http # Alternatively, using uv directly uv run --with "mcp[cli]" mcp run src/linkedin_mcp_server/main.py --transport streamable-http # Go to http://127.0.0.1:10000/mcp ``` The server will start at `http://localhost:10000`. It will automatically reload if you make changes to files in the `src/` directory. ### MCP Inspection Mode ```bash # Using pixi DANGEROUSLY_OMIT_AUTH=true npx @modelcontextprotocol/inspector pixi run mcps --transport stdio # Direct execution DANGEROUSLY_OMIT_AUTH=true npx @modelcontextprotocol/inspector pixi run python src/linkedin_mcp_server/main.py --transport streamable-http ``` This starts the inspector for the MCP Server. ### Web scrapper ```sh pixi run python src/linkedin_mcp_server/web_scrapper.py ``` ## Development Tasks ### Run Tests ```bash pixi run test ``` ### Lint and Check Formatting ```bash pixi run lint ``` ### Apply Formatting and Fix Lint Issues ```bash pixi run format ``` ### Build the Python Package Creates sdist and wheel in `dist/`: ```bash pixi run build ``` ### Docker Support (Optional) #### Build the Docker Image ```bash docker build -t linkedin-mcp-server . ``` #### Run the Docker Container TODO: Rewrite this if necessary. Docker support not yet done. ## MCP Server Installation ### Local Installation with MCP #### Init the mcpb Init mcpb project with a manifest ```sh npx @anthropic-ai/mcpb init --yes ``` **Note** When creating the manifest, in the `mcp_config` section, put the full path to the python interpreter -> `"command": "/Users/fperez/.pyenv/shims/python"` #### Bundle Python libs and Package Project as mcpb ```sh pixi install pixi run mcp-bundle pixi run pack ``` The output file `linkedin-mcp-fps.mcpb` is created on the `mcpb-package` directory. Alternatively, download the `linkedin-mcp-fps.dxt` file from releases (TODO). With the packaged extension: 1. Double-click the `.mcpb` file to install it in Claude Desktop 2. Alternatively, drag and drop it to Claude Desktop Settings/Extensions section 3. Restart Claude Desktop (In new Claude versions it's not necessary) 4. The extension should appear in your MCP servers list ### Extension Requirements This extension requires Python 3.13+ and includes all necessary dependencies bundled. ### Dev Local Installation for Claude Desktop/Code (without DXT) ```json { "linkedin_mcp_fps": { "command": "uv", "args": [ "run", "--with", "mcp[cli]", "--with", "pymupdf4llm", "mcp", "run", "src/linkedin_mcp_server/main.py", "--transport", "streamable-http" ] } } ``` ### Remote Configuration for Claude Desktop/Code For connecting to a remote MCP server: ```json { "linkedin_mcp_fps": { "command": "npx", "args": ["mcp-remote", "http://localhost:10000/mcp"] } } ``` > **Note**: Update the host and port as needed for your deployment. Currently I'm using `render.com` to host the MCP server. The configuration for Claude uses `streamable-http` transport which is the recommended approach for remote deployments. Render requires `requirements.txt` to be present in the root directory. You can generate it using: ```bash uv pip compile pyproject.toml > requirements.txt ``` Also requires `runtime.txt` to be present in the root directory with the Python version specified: ```txt python-3.13.0 ``` Remember also to set the environment variables in the render.com dashboard: ```bash TRANSPORT=streamable-http PORT=10000 ``` ## User Guide ### Available Tools 1. get_url_for_jobs_search: Generate LinkedIn job search URLs 2. get_new_job_ids: Get new job IDs from LinkedIn 3. get_jobs_raw_metadata: Extract detailed job information 4. tailor_cv: Adapt CV to job requirements After installing the MCP server, you can access its functionality in Claude Desktop/Code using the tools to get information about jobs in Linkedin. Combined with the functionality provided by the [MCP Server serving my CV](https://github.com/francisco-perez-sorrosal/cv/tree/mcp) you can ask things like this: ```text Get a list of new jobs from linkedin (2 pages) for a research engineer position in ml/ai in San Francisco, take the last job id from that list, retrieve its metadata, show its content formatted properly, and finally use the tailor_cv_to_job prompt to adapt Francisco's CV to the job's description retrieved. ``` or simply: ```text Use the tailor_cv_to_job prompt to adapt Francisco's CV to the latest job retrieved # or Use the tailor_cv tool and tailor_cv_to_job prompt to adapt Francisco's CV to specific job requirements ``` or, as a recruiter, get your posted LinkedIn job id and write: ```text Use the tailor_cv_to_job prompt to adapt Francisco's CV to this job id 4122691570 ``` ## Cache The system uses a local cache to avoid re-scraping jobs: - Cache location: `~/.linkedin-mcp/raw_job_description_cache/` - Format: JSONL (JSON Lines) - Automatically managed - In-memory cache for fast lookups - Configurable cache keys and storage locations ## Troubleshooting If you encounter issues: 1. **Import errors**: Ensure all required Python packages are installed 2. **WebDriver issues**: Make sure Chrome is installed for Selenium 3. **Connection errors**: Check your internet connection for LinkedIn access 4. **Permission errors**: Ensure the cache directory is writable 5. **Python path issues**: Verify the manifest.json uses the correct Python executable path ## Dev ### Requirements - `pixi` - `uv` ## Support For issues and feature requests, visit: https://github.com/francisco-perez-sorrosal/linkedin-mcp ## License This project is licensed under the MIT License. See `pyproject.toml` (See `LICENSE` file) for details.