obs-mcp

MCP server that lets AI agents send telemetry to explicitly provided endpoints.

Current MVP tools:

1. push_prometheus_metric for Prometheus Pushgateway
2. push_otlp_trace for OTLP HTTP trace export

Each tool requires an endpoint argument. The server does not infer a default endpoint. Each result states which endpoint was targeted.

Stack

1. TypeScript + Node.js
2. @modelcontextprotocol/server
3. zod
4. prom-client
5. OpenTelemetry JS trace exporter

Run

npm install
npm run build
npm run test

Development entrypoint:

npm run dev

Production entrypoint:

node dist/src/index.js

Tools

push_prometheus_metric

Sends one metric sample to the exact Pushgateway endpoint provided by the caller.

Implementation:

1. Tool registration: src/server.ts
2. Handler: src/tools/push-prometheus-metric.ts
3. Validation: src/schemas/prometheus.ts
4. Sink service: src/services/prometheus-pushgateway.ts

Important inputs:

1. endpoint
2. jobName
3. metricName
4. metricType
5. value
6. help

push_otlp_trace

Exports one span to the exact OTLP traces endpoint provided by the caller.

Implementation:

1. Tool registration: src/server.ts
2. Handler: src/tools/push-otlp-trace.ts
3. Validation: src/schemas/otlp.ts
4. Sink service: src/services/otlp-trace-export.ts

Important inputs:

1. endpoint
2. serviceName
3. spanName
4. attributes
5. durationMs or explicit timestamps

Configuration

Optional environment variables are parsed in src/config.ts.

1. SERVER_NAME
2. SERVER_VERSION
3. METRIC_TIMEOUT_MS
4. TRACE_TIMEOUT_MS
5. MAX_LABELS
6. MAX_ATTRIBUTES
7. MAX_STRING_LENGTH

Current Constraints

1. No auth yet
2. No batching
3. No OTLP metrics tool yet
4. stdio transport only