NVIDIA / NVIDIA/NeMo-Agent-Toolkit
Docs: document nat serve /health endpoint in REST API guide
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 2.6k
- Forks
- 762
- Avg merge
- 21h 28m
- Merged PRs (30d)
- 27
Description
Problem
nat serve now exposes a FastAPI GET /health endpoint, but the public REST API server endpoints guide does not mention it. The guide lists the workflow/chat routes and shows how to start nat serve, but an operator looking for a liveness/readiness URL has no documented route to use.
Self-hosted/operator use case
When running NeMo Agent Toolkit workflows behind a reverse proxy, process supervisor, Docker healthcheck, or Kubernetes-style readiness/liveness probe, I need a stable documented endpoint that can be checked without sending an actual workflow request or requiring model/API credentials.
The MCP and FastMCP server guides already document their health checks, so the FastAPI nat serve path feels like the missing piece for production deployment docs.
Evidence inspected
- Current published REST API server endpoints guide: https://docs.nvidia.com/nemo/agent-toolkit/latest/reference/rest-api/api-server-endpoints.html
- Documents
/v1/workflow,/v1/workflow/stream,/v1/workflow/full,/v1/workflow/async,/v1/chat,/v1/chat/stream,/v1/chat/completions, and legacy paths. - Does not mention
/health, liveness, or readiness.
- Documents
- Local source on
develop:packages/nvidia_nat_core/src/nat/front_ends/fastapi/routes/health.py- Registers
GET /health. - Returns
{"status": "healthy"}with HTTP 200. - Description says
Health check endpoint for liveness/readiness probes.
- Registers
- Merged PR that added the endpoint: https://github.com/NVIDIA/NeMo-Agent-Toolkit/pull/1466
- PR body says:
Add GET /health endpoint to nat serve for liveness/readiness probes. Returns {"status": "healthy"} with 200 status code.
- PR body says:
- MCP health docs already exist:
- https://docs.nvidia.com/nemo/agent-toolkit/latest/run-workflows/mcp-server.html documents
/healthandnat mcp client ping. - https://docs.nvidia.com/nemo/agent-toolkit/latest/run-workflows/fastmcp-server.html documents
/healthandnat mcp client ping.
- https://docs.nvidia.com/nemo/agent-toolkit/latest/run-workflows/mcp-server.html documents
Expected behavior
The REST API server endpoints documentation should include the FastAPI server health route, probably near "Default Endpoint Paths" or "Start the NeMo Agent Toolkit Server":
curl -s http://localhost:8000/health
Expected response:
{"status":"healthy"}
It would also help to state that this is intended for liveness/readiness probes and does not execute a workflow.
Suggested implementation shape
Small docs-only update:
- Add
GET /healthtodocs/source/reference/rest-api/api-server-endpoints.md. - Mention that it returns HTTP 200 with
{"status":"healthy"}. - Cross-link, if desired, from the serving/deployment section or REST endpoint table.
- Optionally add an explicit note that MCP/FastMCP have separate health-check docs and default ports.
Duplicate search performed
I searched open and recent closed issues/PRs before filing:
- Pulled 24 open issues, 200 closed issues, 20 open PRs, and 200 closed PRs with
gh. - Searched GitHub for:
health endpoint,healthcheck,readiness probe,liveness probe,nat serve health,MCP health,FastMCP health,deployment health,monitoring health,reverse proxy health,Kubernetes health,docker health, andproduction health. - Relevant hits found:
- #1466 added the FastAPI
/healthendpoint, but did not add it to the current REST API endpoints guide. - #576 added MCP
/healthdocs/behavior. - #1539 added FastMCP support and current FastMCP health docs.
- #1466 added the FastAPI
- I did not find an open documentation issue covering the missing
nat serve/healthdocumentation.
Willingness to contribute
Happy to open a small docs PR for this if maintainers agree the REST API endpoints guide is the right place.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.