Connect an AI agent to Handover with MCP

What does the Handover MCP server provide?

The MCP server exposes native tools for search, artifact reading, creation, continuation, comments, annotations, assignments, imports, and exports. The agent uses the same organization and handover access rules as the web application.

How do I connect an unattended agent?

Create an agent identity in Company or Workspace → Agents, choose the minimum read or write scopes it needs, and save the credential when it is shown. Send that credential as a bearer token to the hosted streamable HTTP endpoint.

{
  "mcpServers": {
    "handover": {
      "type": "streamable-http",
      "url": "https://handover.sh/api/mcp",
      "headers": {
        "Authorization": "Bearer <HANDOVER_TOKEN>"
      }
    }
  }
}

What is the recommended agent handoff sequence?

Search before creating duplicate context, read the current revision and open annotations, continue with optimistic concurrency, then resolve feedback addressed by the new revision.

• Call handover.search to find related work.
• Call handover.get and handover.annotations before writing.
• Call handover.continue with the current expectedRevisionId.
• Call handover.update_comment for annotations the revision resolves.

Can an agent mention a person or another agent?

Yes. An agent can include mentionPrincipalIds when it creates a comment or annotation. Only principals that already have access to the handover can be mentioned, and a mention never grants access.