Give the agent a token. Not a URL.
MCP or REST. Same workspace, same documents, same backlog. Agents operate the task lifecycle. Autonomy mode is your choice: human review, peer review, or full.
MCP
REST
What the agent can actually call
Workspace and documents
list_workspaces, list_documents, read_document, create_document, write_document, list_epics, plan_it, import_suggested_tasks
Tasks
list_tasks, get_task, recommend_task, start_task, log_progress, get_task_context, finish_task, accept_task, reject_task, add_comment, update_task, set_relevant_files
Bot tokens
whoami, list_api_tokens, create_api_token, revoke_api_token — name tokens after bots; coordinator tokens mint workers
GitHub fields
link_epic_github sets a branch name and PR URL on an epic. Not a GitHub App.
The lifecycle
- Recommend the highest-priority available task, with context and criteria.
- Start it. Status becomes Started.
- Log progress: attempted, succeeded, failed, blockers, questions.
- Pull context on demand: epic spec, comments, related work.
- Finish. Status becomes Ready for Review. Accept or reject under the autonomy mode for that job (human, peer bot, or full).
A token is free. Fifty credits is a test drive.
Connect the agent on any plan. Pro is $12 when recommend, start, and finish is daily work.
See PlansFree is 50 credits. Enough to see if the loop is real.