Observability Agent MCP
The Observability Agent MCP is a read-only server that provides AI agents with bounded, structured observability evidence from metrics, alerts, health checks, and Grafana dashboards — without executing commands or modifying systems.
What you can do:
Check capabilities (
observability.capabilities): Discover which providers and tools are enabled, along with configured limits and feature flags.Get service health snapshots (
observability.health_snapshot): Retrieve bounded health status (healthy/degraded/unhealthy/unknown) for named logical services, with summaries and timestamps.List active alerts (
observability.active_alerts): Fetch normalized alert metadata filtered by service, severity (critical/warning/info), and state (firing/pending/resolved), including annotations like summaries, descriptions, and runbook references.Query metrics (
observability.query_metrics): Run bounded instant or range queries using named templates against a Prometheus-compatible backend (including VictoriaMetrics), returning time-series data with labels and samples.Render a Grafana panel as PNG (
observability.render_panel): Fetch an allowlisted Grafana panel as a PNG image for a specified time range, theme, and dimensions, with metadata including byte size, SHA-256 hash, and render duration.Render a Grafana dashboard as PNG (
observability.render_dashboard): Fetch an entire allowlisted Grafana dashboard as a PNG image with the same structured metadata.Build incident context (
observability.incident_context): Assemble a compact evidence bundle for a specific alert or service, combining health data, active alerts, dashboard references, and optionally rendered visuals.
Key constraints:
All tools are read-only — no writes, silences, or remediations
Queries use named templates only — no raw PromQL from the client
Strict limits enforced (e.g., max 50 series, max 100 alerts, max 25 services)
All outputs include freshness indicators, truncation flags, and redaction notices
Visual rendering is opt-in and restricted to an allowlist of dashboards/panels
Consumable via stdio, authenticated Streamable HTTP, OCI image, or npm CLI
Provides tools for querying metrics, rendering panels and dashboards as PNG evidence, retrieving active alerts, and building compact incident context bundles from a Grafana instance, with read-only access and deterministic limits.
Planned direct integration for querying metrics from Prometheus-compatible backends, providing bounded instant and range queries with normalized evidence bundles, label redaction, and safe limits.
Planned direct integration as a Prometheus-compatible backend, allowing the same bounded metrics query and evidence contract as Prometheus without requiring Grafana.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Observability Agent MCPshow me the current health of all services"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Observability Agent MCP
A portable, read-only Model Context Protocol server for turning observability data into bounded evidence that AI agents can inspect safely.
Status
Runnable provider-backed MCP. The TypeScript server exposes seven bounded, read-only tools over stdio and authenticated Streamable HTTP. It includes VictoriaMetrics/Prometheus-compatible metrics and alert adapters, allowlisted Grafana PNG rendering, deterministic local fixtures, npm packaging, and a non-root OCI runtime. Private deployment configuration remains outside this public repository.
Related MCP server: Observe MCP Server
Product Boundary
This project provides deterministic tools for observability evidence. It does not run an LLM, receive chat messages, execute shell commands, deploy workloads, or remediate production systems.
AI client
| stdio (implemented)
v
Observability Agent MCP
- strict tool schemas
- query and response limits
- label and value redaction
- normalized evidence bundles
|
+--> deterministic fake provider
+--> allowlisted Grafana render API
+--> named-query Prometheus-compatible API
+--> VictoriaMetrics alert API
- Prometheus
- VictoriaMetricsThe model and agent loop remain in the client. This server only returns structured evidence.
First Tool Set
Tool | Purpose |
| Describe configured providers and safe limits. |
| Return bounded service and target health. |
| Return normalized active alert metadata. |
| Run a bounded instant or range metrics query. |
| Return one allowlisted Grafana panel as PNG evidence. |
| Return one allowlisted Grafana dashboard as PNG evidence. |
| Build a compact evidence bundle for one alert or service. |
Version 1 is read-only. It will not create dashboards, modify alert rules, silence alerts, restart services, run scripts, or trigger deployments.
Portability
The server can be consumed through:
stdio for local desktop and CLI agents;
an authenticated Streamable HTTP endpoint on a private network;
a non-root OCI image;
the package CLI produced by
npm pack.
The implemented HTTP mode uses a file-injected Bearer credential and a strict Host allowlist for a private, single-operator deployment. It is not a public or multi-tenant OAuth deployment. No package or image has been published yet.
Remote HTTP deployment requires authentication. Publishing this repository does not imply that an MCP endpoint should be exposed without OAuth, network policy, rate limits, and audience-bound scopes.
Run Locally
Node.js 22 and npm are required.
npm ci
npm run build
node dist/cli.jsThe process speaks MCP over stdio, so launch it from an MCP client rather than typing into the terminal. A client command definition is:
{
"command": "node",
"args": ["/absolute/path/to/observability-agent-mcp/dist/cli.js"]
}Inspect tool discovery and run every local gate with:
npm run inspector:list
npm run validateBuild and run the non-root OCI image over stdio with:
npm run container:build
docker run --rm -i observability-agent-mcp:localProduction HTTP mode requires a strict YAML provider policy and two 0600
runtime credential files. See Client Compatibility
and Security Model before deploying it.
Relationship To Grafana MCP
The official Grafana MCP server is the preferred reference and fast path for broad Grafana capabilities. This project will not duplicate its dashboard and administration surface.
Capability | Grafana MCP | This project |
Broad Grafana dashboards and administration | Primary owner | Not implemented |
Grafana datasource queries | Supported | Narrow adapter only |
Direct Prometheus-compatible backend without Grafana | Not the primary path | Supported goal |
Provider-neutral incident evidence | Provider-specific | Primary owner |
Conservative read-only public contract | Configurable broad surface | Required default |
The unique scope here is:
compact incident-context evidence;
conservative read-only defaults;
direct Prometheus-compatible operation when Grafana is absent;
provider-neutral schemas;
predictable limits and redaction for agent use.
Grafana visuals are first-class evidence. Rendering tools return an MCP
ImageContent PNG plus structured metadata for the source, observation window,
dimensions, byte size, freshness, truncation, and warnings. Visuals are opt-in;
normal metric queries do not spend rendering resources or image context.
There is no automatic delegation between the two servers. Use Grafana MCP when
an agent needs Grafana-native dashboards, incidents, administration, or its
existing query tools. Use this project's namespace when the client needs the
same bounded evidence contract across Grafana and direct
Prometheus-compatible backends. active_alerts and incident_context are
normalized here rather than forwarded to Grafana MCP; query_metrics selects
one configured adapter and never calls another MCP server. Deployments may
offer both namespaces, but each request has one explicit owner.
Read Next
License
Apache License 2.0. See LICENSE.
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/hamardikan/observability-agent-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server