MCP reference

Tool examples and access details. Connect your agent →

Read project tasks

Replace placeholder IDs with real IDs. Current tools accept flat arguments. The connected server’s advertised schema is authoritative; older deployments may advertise nested arguments.

task_list

List project tasks across assignees, with a small response.

{
  "projectId": "<projectId>",
  "fields": "id,name,status",
  "limit": 25,
  "page": 1
}

task_search

Find matching tasks within a project.

{
  "term": "billing",
  "projectId": "<projectId>",
  "fields": "id,name,status",
  "limit": 25,
  "page": 1
}

task_get

Read one task before acting on it.

{
  "taskId": "<taskId>"
}

Most results use a status and data envelope. Check isError and individual batch results before assuming success; status 207 means partial success. For mutations, choose a requestKey before sending and reuse it with identical arguments when retrying a lost response.

Access and agent workflows

Each connection is tied to one user and account. Requested scopes separate read and write access for work, integrations, and account operations; your existing roles and project membership still apply. Authorize a new connection to switch accounts. Revoke access in Profile → Integrations or with p3 auth connections list/revoke.

Agents can read context, claim work, record progress, and track linked PRs through the tools advertised by the server. A live claim and recorded progress are separate signals. Task status and delivery are also separate. Local commands, repository checkout, and running verification checks remain the client’s responsibility.

Follow the CLI agent workflow or install the TaskP3 skill for detailed agent guidance.