Mcp
MCP run dispatch
Dispatch and inspect agent runs via MCP capabilities.
MCP run dispatch
External MCP clients dispatch agent runs through capability tools
that share the same dispatch pipeline as POST /v1/runs.
Tools
| Tool | Purpose |
|---|---|
cap.run.dispatch | Dispatch a run by agent slug |
cap.run.get | Fetch run status; optional steps and artifacts |
cap.approval.resolve | Approve or reject a pending approval gate |
cap.run.dispatch
Input:
| Field | Required | Description |
|---|---|---|
agent_slug | yes | Agent slug in the active workspace |
input | yes | JSON object passed as run goal input |
conversation_id | no | Link to an existing conversation |
idempotency_key | no | Dedup key (≤200 chars) |
budget_usd | no | Per-run USD cap |
cost_strategy | no | Model routing strategy |
Output includes run_id, execution_id, status, and
dispatched_via (mcp).
cap.run.get
Input:
| Field | Required | Description |
|---|---|---|
run_id | yes | Run UUID |
include_steps | no | Include step list (default false) |
include_artifacts | no | Include artifact list (default false) |
Output:
| Field | When | Description |
|---|---|---|
run.id | always | Run UUID |
run.status | always | Run status |
run.execution_id | always | Execution UUID |
run.agent_id | always | Agent UUID |
run.started_at | always | ISO-8601 start time, or null |
run.ended_at | always | ISO-8601 end time, or null |
run.step_count | always | Total step count |
steps | include_steps=true | Array of {seq, kind, started_at, ended_at, this_hash} |
artifacts | include_artifacts=true | Array of {id, kind, name} |
cap.approval.resolve
Input: approval_id, decision (approved | rejected), optional
note. Returns the updated approval and run summary.