Retasc MCP — AI Agent Task Orchestration for Hermes Agent¶
Retasc is the MCP server AI coding agents pull work from. It replaces traditional issue trackers with atomic claims, dependency-aware dispatch, and parallel swarm execution — purpose-built for agentic workflows, not human project management.
What It Does¶
Retasc transforms how AI agents consume and complete work:
- Atomic claims with lease/TTL — Agents claim individual work items with per-claim tokens and expiration. No double-work, no orphaned tasks.
- next_batch wave dispatch — Parallel agent swarms pull the next ready batch based on dependency resolution and effective priority.
- Dependency-graph scheduling — Tasks blocked on incomplete dependencies never surface. When a blocker completes, dependents become available automatically.
- Resumable handoffs — Any runtime (Claude Code, Codex, Cursor) resumes from the last checkpoint. Cross-runtime handoffs are native, not bolted on.
- Deadline-aware dispatch —
dueAtSLA pressure: a breach outranks an urgent-priority task. - Inbound intake webhooks — GitHub and GitLab issues sync in; done-items sync back to the source.
- Outbound notifications — Slack, Discord, and Telegram DMs on done, assigned, review, or canceled.
- Human principal binding — Every claim and edit is traceable to the signed-in human who authorized the agent.
Why It Matters for Operators¶
If you're running multiple AI agents on a project, traditional issue trackers (Jira, Linear, GitHub Issues) break down. Agents need atomic work claims, not assignment fields. They need dependency-aware dispatch, not manual backlog grooming. Retasc is the first MCP-native work queue built for how agents actually operate.
Quick Setup¶
Prerequisites¶
- Retasc account: Sign up at retasc.com
- MCP-compatible client: Claude Code, Codex, Cursor, or any HTTP MCP client
Connection Details¶
| Field | Value |
|---|---|
| Transport | Streamable HTTP (Remote) |
| Endpoint | https://api.retasc.com/mcp |
| Authentication | API key (generated in Retasc dashboard) |
| Tools | 33 (claim, release, complete, next_batch, create_issue, dependencies, notifications, etc.) |
Add to Hermes Agent¶
{
"mcpServers": {
"retasc": {
"transport": "http",
"url": "https://api.retasc.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_RETASC_API_KEY"
}
}
}
}
Claude Code Setup¶
claude mcp add retasc --transport http https://api.retasc.com/mcp
# First connection prompts for API key
Key Tools¶
| Tool | Purpose |
|---|---|
claim_issue |
Atomically claim next available work item with lease TTL |
release_issue |
Release a claim (agent blocked, context lost, reprioritized) |
complete_issue |
Mark work done; triggers dependency resolution for dependents |
next_batch |
Fetch next batch of claimable items based on priority + dependency graph |
create_issue |
Create new work items with dependencies, deadlines, and priority |
add_dependency |
Link an issue as blocked on another |
get_status |
Query current state of any issue or batch |
notify |
Send human notifications (Slack, Discord, Telegram) on status changes |
Use Cases for CorpusIQ¶
Agent Swarm Product Development¶
Feed Retasc with feature specs. Multiple Claude Code agents claim tasks in parallel, each with a lease. When one finishes, its dependents unlock automatically. You review PRs while agents orchestrate themselves.
Content Pipeline with Deadlines¶
Create content tasks with dueAt timestamps. Retasc's deadline-aware dispatch ensures time-sensitive posts surface above non-urgent work. Agents check next_batch and always pull what matters most.
Cross-runtime Handoffs¶
Start a research task in Claude Code, checkpoint it. Resume from Codex or Cursor. Retasc handles the context handoff — same claim, same state, different runtime.
Pricing¶
- Free tier: First $10 of metered usage included
- Pay-per-action: After free tier; Retasc reports their own busiest agent runs ~$9/month
- No per-seat: Pay for what agents consume, not how many humans have accounts
Limitations¶
- MCP-native only: No traditional web UI for human task management — this is purpose-built for agents
- Early stage: New product, ecosystem integrations growing
- API key auth: Not OAuth; key management is manual
- Dependency on MCP client support: Requires clients that support Streamable HTTP transport
Verdict¶
Retasc is the most significant agent orchestration MCP we've seen this month. It solves the fundamental mismatch between human-centric issue trackers and agent workflows. For operators running multiple AI coding agents, this replaces the Jira/Linear API + custom dispatch scripts pattern with a single MCP endpoint. Strong recommend for any multi-agent workflow.
Rating: ★★★ — Essential for multi-agent operators