Give your own assistant a verb set over your funnel: it can read your pending candidates (the same numbers as the funnel chips), record what you already did on LinkedIn, move cards the way a board drag does, and flag your screened list to the AM. Everything it does is logged under your name, with your permissions — nothing more.
Claude is what we use and what the setup below is written for; any assistant that speaks MCP can connect.
Add this URL as a custom connector in your AI assistant — you’ll be asked to sign in and approve. There is no token to copy and no header to configure.
In the Claude app: Settings → Connectors → Add custom connector, paste the URL, and press Connect. A sourcing-app sign-in opens; approve it and you’re done.
Approving connects that one assistant to your account. It acts as you, with your permissions, and every change carries your name.
Loading…
Disconnecting takes effect on the assistant’s very next call — there is nothing left for it to refresh with.
No money, no stamps, no role/portal administration, no other user’s data, no submissions (submitting stays the AM’s decision — your assistant can only put your screened list in front of them), and no LinkedIn sending. Refusals explain themselves.
Only if your assistant can’t do the sign-in above — some tools speak MCP but not OAuth. This mints a private link with a credential inside it, which you paste as the connector URL instead.
The same credential authenticates as a bearer token against the bare endpoint. In Claude Code:
claude mcp add --transport http sourcing <URL>
A credential minted above is what goes in the header — but it is only ever shown inside the link, so copy it from there before you close this page.
Then just talk: “what’s pending on my replied stage?” · “I messaged these 12 on LinkedIn, mark them contacted” · “she said not now — log it, re-engage in October”. Your assistant confirms before batch writes; LinkedIn sending stays in your own LinkedIn tools — this connection only records outcomes.