LinkedIn for AI agents: MCP servers, hosted APIs and Relay.
Three ways to give an agent LinkedIn access. Each has trade-offs.
Free local MCP servers
A widely used one is stickerdaniel/linkedin-mcp-server (3.9k GitHub stars as of October 2026). It is open source (Apache 2.0) and free. It runs on your computer and drives LinkedIn with Patchright browser automation, using your own session. It has more than 20 tools, including profiles, companies, jobs, the feed and messages. [1]
If it does what you need, it is a good choice.
Hosted LinkedIn APIs
Services such as Unipile connect your account to their API and charge per connected account each month. They are built for software publishers. [2] See Relay vs Unipile.
Relay
- A Node.js 22+ CLI and local HTTP API. It works with any agent that has a shell today: Claude Code, Codex, Pi and others.
- Your agent calls it from its shell, and gets results on average 10x faster than an agent clicking through LinkedIn in a browser.
- LinkedIn runs in a browser on your computer, signed in as you. Relay's cloud never sees LinkedIn content.
- A fixed operation contract with JSON schemas. Each result says what is missing or partial.
- Message sends return a receipt with the verified recipient and exact text. An unclear send is never retried.
- $0.01 per completed request. No subscription. $5 free credit covers 500 requests.
Relay also has an optional relay mcp adapter for MCP clients such as Claude Desktop, Claude Code, Cursor and Codex. It is available now, from release 0.3.0. See MCP clients and every operation in the docs.
Side by side
| Feature | Free MCP server | Hosted API | Relay |
|---|---|---|---|
| Runs | Your computer | Provider's servers | Your computer |
| Agent connects by | MCP | REST API | Shell CLI, local HTTP, or MCP (optional adapter) |
| Price | Free | Monthly, per account | $0.01 per completed request |
Risks
Every option here automates LinkedIn. Relay is not affiliated with LinkedIn. Automation may breach LinkedIn's User Agreement (section 8.2). Use your own account, at a normal pace.
Start
Sources
Checked on 17 October 2026.