AI agents for BullMQ, over MCP
Connect Claude, Cursor or your own agent to your BullMQ queues through the Taskforce.sh MCP server. Agents can investigate failed jobs, retry the ones that are safe and pause a queue, limited by the same permissions you give a teammate, and every action is logged.
How it works
1. Add an agent
Create an agent in your organization and add it to a team with a role, just like inviting a person. It only sees that team's queues.
2. Create a token
Give the agent a full or read-only token, optionally with an expiry date.
3. Connect your client
Point your MCP client at https://api.taskforce.sh/mcp with the token as a Bearer credential.
What agents can do
Investigate
- List queues and their job counts and metrics
- List failed jobs with their failure reasons
- Read a job's data, options, logs and flow
- List job schedulers
Act
- Retry or promote jobs, one by one or in batches
- Add jobs
- Pause and resume queues
- Remove jobs and clean queues, with a dry run first
See every tool in the tools reference.
Guardrails
Roles and teams
Agents get permissions through team roles, including custom roles. An operator role, for example, can retry and promote jobs but not delete them.
Read-only tokens
A read-only token can never call a write tool, whatever the agent's role.
Audit log
Every tool call is recorded with its outcome, allowed or denied. Export the log when you need a record.
MCP questions
Any MCP client. Taskforce.sh exposes a standard Streamable HTTP endpoint that works with Claude Code, Cursor, VS Code and agents built with an MCP SDK. Clients that only run local servers, such as Claude Desktop, connect through the mcp-remote bridge.
https://api.taskforce.sh/mcp. Agents authenticate with a Bearer token that you create for them in Taskforce.sh.
Only if you allow it. An agent can only take the actions its role grants on each team, and read-only tokens cannot call any write tool. Destructive tools such as removing jobs or cleaning a queue support a dry run.
Only the queues on connections that belong to the teams the agent is a member of. Personal connections are never visible to agents.
Each organization can have up to three agents. Each agent gets its own tokens, which can be read-only and can expire.
Yes. Every tool call is recorded in the audit log with its outcome, allowed or denied, and the log can be exported.
Let an agent triage your queues
Create an agent, give it a read-only token and see what it finds.
15-day free trial · No credit card required
