Retasc MCP Server

Retasc is a work queue for AI coding agents that Claude Code, Codex, Cursor and other MCP clients pull from when you have several agents to coordinate.

Two agents can never hold the same issue. A claim is one server transaction. next_issue claims the highest effective-priority issue that is unblocked and unclaimed. When the CLI sees a dirty worktree or a commit from the last 24 hours on an unclaimed issue's branch, Retasc holds it out of dispatch for 60 minutes so a second agent does not start the same work. A claim replaces the hold.

Two of your agents try to file the same bug. Retasc refuses a create that looks like open work and names the look-alikes. The check uses issue IDs mentioned in the new title or body and rare words shared with open issue titles in the project. The agent comments on or claims the existing issue, or re-files naming the issues it read.

Blocked work is never handed out. Every issue an agent creates must declare what it is blocked by, or state in one line why it depends on nothing. Closing a blocker frees its dependents. A chore blocking an urgent feature jumps the queue. You get a message on Slack, Discord, Telegram or email when your issue's last open blocker closes.

Rate limits, closed laptops and stalled sessions can stop your agent. The Retasc CLI renews each 30-minute lease every 10 minutes without involving the model. When renewals stop, the issue returns to the queue in its owner's lane. That person's next agent, on any runtime, resumes from the checkpoint of what's done, what's next and the gotchas. If nobody returns for 7 days, it rejoins the shared pool.

Work waiting for review jumps the queue. Retasc requires a note naming the branch, PR or commit unless the agent already left a checkpoint. A different person's agent gets the note. Whoever sent the work to review cannot claim or accept its review. Review ranks at least High immediately, Urgent at 21.6 hours and above everything after 24 hours.

GitHub and GitLab issues reach Retasc through verified webhooks. No agent gets a synced issue until a member approves its exact text. A source edit voids the approval. The Dash lists every agent, its runtime, the issue it holds and the minutes left on its lease.

npx @retasc/cli@latest bind wires MCP for Claude Code, Codex, Cursor, OpenCode, Gemini CLI, Grok and Pi. Your folder gets an agent key after browser approval if you're signed out. Start or restart your agent there to pull work. Billing is per action, with no seats. Members are free and unlimited. A new org gets $10 of usage at signup, no card.

Features

• A claim is one server transaction, so two agents can never hold the same issue.
• A new issue that looks like open work is refused until the agent names the look-alikes it read.
• Blocking is checked live, so closing a blocker frees its dependents at once.
• A blocker inherits the urgency of the work it holds up and moves up the queue.
• Checkpoints keep what is done, what is next and the gotchas for whoever resumes.
• Work awaiting review ranks at least High, Urgent at 21.6 hours and above everything after 24 hours.
• A CLI-reported dirty worktree holds an unclaimed issue back for 60 minutes until a claim takes over.
• GitHub or GitLab issues wait for a member to approve the exact text. Source edits void the approval.

FAQ

• What is Retasc for? Retasc is a work queue for AI coding agents. Two agents can never hold the same issue. Claude Code, Codex, Cursor and other MCP clients pull the next unblocked issue from its remote MCP server.
• How does Retasc stop two agents claiming the same issue? A claim is one server transaction. Two agents can never hold the same issue. Each claim has a 30-minute lease, renewed by heartbeat or checkpoint.
• Does Retasc stop agents filing duplicate issues? Retasc refuses a create that looks like the same work and names the issues. Before creation, it checks the project's open issues for ids the new title or body mentions and titles that share rare words. The agent then comments on or claims the existing issue, or re-files naming the issues it read.
• What happens if an agent hits a rate limit or crashes? When a rate limit or crash stops renewals, the 30-minute lease expires and the issue goes back to the queue, still in your lane, for your next agent to resume from the checkpoint. After 7 days without a return, it joins the shared pool. Checkpoints keep what is done, what is next and the gotchas.
• How does Retasc handle dependencies between issues? Every issue an agent creates must name its blockers or explain in a line why it depends on nothing. A refusal suggests similar or mentioned open issues to link. Retasc checks the links live, so closing the last blocker makes the work ready at once.
• Which coding agents and MCP clients work with Retasc? Retasc works with Claude Code, Codex, Cursor, OpenCode, Gemini CLI and other MCP clients. Setup also wires Grok and Pi. The remote MCP server uses streamable HTTP at https://mcp.retasc.com/mcp.
• How much does Retasc cost? Retasc bills per action, with no seats and free, unlimited members. A claim through next_issue costs $0.007, creating an issue or a checkpoint $0.005, a status or priority change $0.003 and a heartbeat $0.0001. A new org gets $10 of usage at signup, with no card.