An open-source AI SDR for LinkedIn
Last updated 19 September 2026
HeyLead is an AI SDR for LinkedIn, and the code is on GitHub under the MIT licence. It finds the right people, writes to them in your voice, follows up and handles replies. You can run it yourself or use the hosted plan.
Source: github.com/D4umak/linkedin-outreach-mcp. Package: heylead on PyPI.
What is in the repo
- The MCP server, with 35 tools for your AI client (Claude Code, Claude Desktop, Cursor, any stdio MCP client)
- Targeting: buyer personas from a plain description, LinkedIn search and fit scoring
- Writing: voice matching from your own LinkedIn posts, openers and follow-ups
- Replies: reading and answering replies, meeting booking
- A local scheduler with rate limits and quiet hours
Self-host it
You need two accounts of your own:
- Unipile, which connects to LinkedIn. Sign up at unipile.com and put the DSN and API key in
~/.heylead/config.jsonasunipile_api_urlandunipile_api_key. - An LLM API key from Gemini, Claude or OpenAI. AI calls are billed to you; a free Gemini key is enough to start.
claude mcp add heylead -- uvx heylead
Then tell your AI: "Set up my HeyLead profile with this Gemini key: ...". Contacts and messages stay on your machine in a local SQLite database. Sending runs from your machine, so it stops when the machine sleeps.
Or use the hosted plan
Sign in at heylead.dev, connect LinkedIn and paste the setup message into your chat. No Unipile account and no API keys. Campaigns run from the cloud around the clock, and there is a web dashboard next to the chat.
How it reaches LinkedIn
Through Unipile's API. It does not use a Chrome extension or drive your browser. LinkedIn's User Agreement prohibits automated invitations and messages, so any tool like this carries a risk to the account. HeyLead keeps to daily and weekly limits and a weekday sending window, which lowers the risk but does not remove it.