GlassFlow MCP Server

Model Context Protocol server for managing and diagnosing GlassFlow streaming pipelines. Exposes pipeline CRUD, metrics queries, log search, and a composite diagnostic tool as MCP tools that AI agents (Claude Code, etc.) can call over SSE transport.

Features

• Multi-cluster — connect to multiple GlassFlow deployments and switch between them at runtime

• Pipeline management — create, list, get, edit, stop, resume, delete pipelines

• Diagnostics — query throughput, latency, DLQ state, and error logs

• NATS JetStream diagnostics (Enterprise) — inspect stream message counts and consumer health (get_pipeline_streams, get_stream_report, get_consumer_report) to spot stuck consumers, backlogs, and subject mismatches

• diagnose_pipeline — single-call diagnostic snapshot combining health, metrics, DLQ, recent errors, and NATS stream health

• V3 config reference — MCP resource with the complete pipeline configuration format

• Uses the official GlassFlow Python SDK

Related MCP server: Skemati.ca MCP SSE Server

Quick start

Local development

pip install -e .

# Option A: auto-connect a default cluster via env var
export GLASSFLOW_API_URL="http://localhost:8081"
python -m glassflow_mcp.server

# Option B: start with no cluster, connect at runtime via tools
python -m glassflow_mcp.server

Connect Claude Code

claude mcp add --transport sse glassflow http://localhost:8080/sse

Start a new Claude Code session — the GlassFlow tools will appear automatically.

Available tools

Cluster management

Connect to one or more GlassFlow clusters and switch between them. All pipeline and diagnostic tools operate against the active cluster.

Example flow:

You: "Connect to my staging cluster at http://staging-api:8081"
  → Agent calls: connect_cluster(name="staging", api_url="http://staging-api:8081")

You: "List my pipelines"
  → Agent calls: list_pipelines()  (uses staging)

You: "Now connect to production at http://prod-api:8081"
  → Agent calls: connect_cluster(name="production", api_url="http://prod-api:8081")

You: "Switch to production"
  → Agent calls: switch_cluster("production")

You: "List pipelines"
  → Agent calls: list_pipelines()  (now uses production)

If GLASSFLOW_API_URL is set as an env var, the server auto-connects a default cluster on startup for backwards compatibility.

Pipeline management

Diagnostics

Resources

Configuration

All configuration is via environment variables. These configure the default cluster that auto-connects on startup. Additional clusters can be added at runtime via connect_cluster.

VictoriaMetrics and VictoriaLogs URLs are optional — metrics and log tools gracefully degrade when not configured for a cluster.

Deployment

Docker

docker build -t glassflow-mcp-server .
docker run -p 8080:8080 \
  -e GLASSFLOW_API_URL=http://your-glassflow-api:8081 \
  glassflow-mcp-server

Kubernetes

Example manifests are provided in k8s/examples/. Copy them, edit the CHANGEME values, and apply:

kubectl apply -f k8s/examples/deployment.yaml -f k8s/examples/service.yaml

Then connect via port-forward:

kubectl port-forward -n <namespace> svc/glassflow-mcp 8080:8080
claude mcp add --transport sse glassflow http://localhost:8080/sse

See k8s/README.md for full details including optional Ingress setup.

PyPI

pip install mcp-server-glassflow
mcp-server-glassflow

Development

# Install with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest -v

# Lint
ruff check src/ tests/
ruff format --check src/ tests/

License

Apache 2.0