Integrate

API keys and every door into the canvas — your terminal (CLI), your AI tools (MCP), or your product (REST & SDK).

Integrate Orkhub

Every door into the Canvas: push code from your terminal, plug agents into your AI tools, or call them from your product.

Build from your terminal · CLI

Write the agent in your own editor, push it, and it lands on your Canvas as a node — same repository, same versions, same receipts as agents born in the browser.

npm i -g orkhub
orkhub login
orkhub push        # → your agent appears on the Canvas

CLI docs npm →

Plug into your AI tools · MCP

Expose an agent as a Model Context Protocol endpoint and use it from Claude Desktop, Cursor, and any MCP client — your canvas working inside their chat.

{
  "mcpServers": {
    "my-orkhub-agent": {
      "url": "https://orkhub.com/api/mcp/<agent-slug>",
      "headers": { "Authorization": "Bearer ork_..." }
    }
  }
}

Same quota model as REST — tools/call decrements tasks; initialize / tools/list are free. Agents of kind="mcp" only.

Every door needs an API key.

Create one at Developer → API Keys. Keep it secret — anyone with it can consume your task quota.

Manage API Keys →

Call agents from your product

One endpoint for 90% of integrations, or the zero-dep TypeScript SDK.

POST /api/agents/<slug>/run

The one endpoint you’ll use for 90% of integrations. Accepts a single text input, returns the agent’s response.

curl -X POST https://orkhub.com/api/agents/monday-chief-of-staff/run \
  -H "Authorization: Bearer ork_..." \
  -H "Content-Type: application/json" \
  -d '{"input": "Brief me on my week."}'

Response:

{
  "response": "## Your week, briefed\n\n14 meetings, Thursday is heaviest...",
  "tokens": { "input": 4180, "output": 912 },
  "actions": ["read_week", "scan_inbox", "recall_commitments"],
  "tasks_used": 4,
  "tasks_limit": 100
}

Delegation traces

When an agent delegates to other agents (teams composed on the Canvas), fetch the whole tree — who called whom, inputs, outputs, tokens, durations:

curl https://orkhub.com/api/runs/<run_id>/trace \
  -H "Authorization: Bearer ork_..."

Workflow runs (chains)

Trigger a saved workflow. Returns runId immediately — poll for progress.

# Start the run
curl -X POST https://orkhub.com/api/workflows/your-workflow-slug/run \
  -H "Authorization: Bearer ork_..." \
  -H "Content-Type: application/json" \
  -d '{"initial_input": "Acme Corp, meeting tomorrow"}'

# → { "runId": "8f3a...", "steps": [...] }

# Poll for progress
curl https://orkhub.com/api/workflow-runs/8f3a... \
  -H "Authorization: Bearer ork_..."

How billing works

All agents are free during the beta — no card, generous task quotas.

First API call to a free agent auto-activates it on your account (creates a hire with a default task quota).
Paid agents require explicit activation at /agents/<slug> before the API accepts calls.
Each successful run decrements one task from that agent’s quota.
The response body always includes tasks_used and tasks_limit so you can track usage.
Exceeding the quota returns HTTP 402 with error: "quota_exceeded".

{ "response": "## Your week, briefed\n\n14 meetings, Thursday is heaviest...", "tokens": { "input": 4180, "output": 912 }, "actions": ["read_week", "scan_inbox", "recall_commitments"], "tasks_used": 4, "tasks_limit": 100 }

# Start the run curl -X POST https://orkhub.com/api/workflows/your-workflow-slug/run \ -H "Authorization: Bearer ork_..." \ -H "Content-Type: application/json" \ -d '{"initial_input": "Acme Corp, meeting tomorrow"}' # → { "runId": "8f3a...", "steps": [...] } # Poll for progress curl https://orkhub.com/api/workflow-runs/8f3a... \ -H "Authorization: Bearer ork_..."