Connect your account
Sign in through hosted Anchor OmniConnect and return to Behalf. Your agent receives a Behalf grant token, rather than your LinkedIn login.
LinkedIn developer preview
Give your agent a task.
Keep the authority.
Let an agent read your LinkedIn conversations and prepare a reply. You choose its scope, approve each message, and inspect what happened.
The demo uses fictional data. Real setup connects your own account.
One clear handoff, from start to receipt.
Set the boundaries
Choose read-only access or replies with approval. Set an expiry and daily limit. Connect your MCP client or use the controlled action API.
Review the exact message
Outbound messages wait for the account owner to approve the saved recipient and text. An agent cannot approve its own request.
Check the result
Inspect the action receipt. It separates a permission decision from an execution attempt and an observed result. An uncertain write is not automatically repeated.
What works in this preview
- LinkedIn through the managed Behalf execution path.
- Remote MCP over Streamable HTTP and the action API.
- Owner-issued grants, approval holds, action receipts, and revocation.
What the boundary means
Behalf governs requests sent through Behalf. It does not control access your agent holds elsewhere. Revoking a grant blocks new dispatches; it cannot undo an action the destination has already accepted.
Read the security modelStart with a read.
Create a short-lived grant and connect your client. Add replies when you are ready to review the recipient and message.
Follow the developer guide →