grafana-mcp-adapter
Provides tools for searching and retrieving Grafana dashboards, listing datasources, running PromQL/LogQL queries, and generating Grafana Explore links for log previews.
Click on "Deploy 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., "@grafana-mcp-adapter Show me the engine logs for Contoso over the last 6 hours."
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.
grafana-mcp-adapter
Read-only MCP server that exposes Grafana dashboards, datasources and metric/log queries as tools, for use by skills in the TICKETS workspace.
This is a private, standalone adapter: its own git repo, its own .env. It is
not part of ia-tooling and does not depend on its stack — it only makes HTTPS
calls out to the Grafana HTTP API (${GRAFANA_BASE_URL}/api). It was scaffolded by
cloning the structure of fathom-mcp-adapter.
Architecture
The adapter is the MCP server. Internally it calls the Grafana HTTP API directly (just as the Fathom adapter calls Fathom's REST API). MCP servers are not chained to one another.
Auth: a Grafana service account token sent as a Bearer token in the
Authorizationheader. Scope it to a read-only role (Viewer). Create the service account + token in Grafana → Administration → Service accounts.Most tools return raw payloads (dashboard JSON, datasource lists). The exceptions are the high-volume ones:
grafana_queryreturns a compact per-series digest by default (a fullupquery is ~8 MB of frames, far past what an MCP context wants), andgrafana_logs_linkreturns a capped preview. Both can be pushed back toward raw (raw=true/ a higherlimit).Read-only by design. Note that
grafana_queryis aPOST(Grafana's/api/ds/queryis POST-shaped) but only reads metrics/logs.
Related MCP server: mcp-grafana
Tools
Tool | Purpose |
| Config/connectivity check (makes one authenticated call). |
| Search dashboards by title substring and/or tags. |
| Full dashboard JSON by |
| List configured datasources (uid, name, type). |
| Run a PromQL/LogQL/etc. query against a datasource uid over a time range. Returns a per-series digest by default. |
| Build a permanent Grafana Explore (Loki) deep link for a customer's logs and return a preview of recent lines. |
grafana_query response shape
The raw /api/ds/query response carries a full timestamp+value array per series,
and a query like up can return thousands of series (~8 MB). By default the tool
collapses each series to its labels + a numeric digest and caps the list:
{
"results": {
"A": {
"status": 200,
"series_count": 3085, // total series before capping
"series": [ // capped to maxSeries (50)
{ "labels": { "job": "..." }, "count": 60, "first": 1, "last": 1, "min": 0, "max": 1, "avg": 0.98 }
],
"truncated": 3035 // how many series were dropped from `series`
}
}
}Pass raw=true to get the full (potentially very large) frames instead.
grafana_logs_link
Identify a customer/component with free text (client='april',
component='gateway'); it matches case-insensitively against the service_name
label, which on this instance encodes both (e.g.
graviteeio-ae-april-rec-engine). Returns { query, explore_url, range, preview_count, preview } — paste explore_url (a permanent Grafana 11+ Explore
link) into the ticket. The default range is the last hour; widen with
from/to. When nothing matches, it returns close service_name values as
suggestions so typos like aprl → april surface.
Examples (how a user asks for it)
Just ask in plain language — the agent maps it to the client / component /
from / to / line_filter arguments for you.
"Give me the last hour of API gateway logs for Northwind." →
{ "client": "northwind", "component": "gateway" }
"Show me the engine logs for Contoso over the last 6 hours." →
{ "client": "contoso", "component": "engine", "from": "now-6h" }
"Find the gateway errors for Globex in the last 3 hours, give me up to 200 lines." →
{ "client": "globex", "component": "gateway", "line_filter": "error", "from": "now-3h", "limit": 200 }
"I need the UI logs for Initech during yesterday's incident between 10:00 and 11:00." →
{ "client": "initech", "component": "ui", "from": "<epoch ms 10:00>", "to": "<epoch ms 11:00>" }
"Give me the production gateway logs for Northwind." →
{ "client": "northwind prod", "component": "gateway" }
(The environment — prod, rec, dev — isn't a separate argument: it lives
inside service_name, so just fold it into client as another word. Both
client and component are matched as case-insensitive substrings with .*
between them, so northwind prod matches …-northwind-prod-….)
Each call returns explore_url — paste it into the ticket as a permanent link —
plus a capped preview of the newest matching lines.
Setup
npm install
cp .env.example .env # then fill in GRAFANA_BASE_URL and GRAFANA_TOKEN
npm test
npm start # speaks MCP over stdioGRAFANA_TOKEN must come from .env (which is git-ignored) — never hardcode it.
Wiring into the TICKETS MCP session
Build the image and add a grafana entry to the TICKETS .mcp.json, alongside
fathom / zendesk / vectordb / github. It only needs HTTPS egress.
docker build -t grafana-mcp-adapter .// .mcp.json (TICKETS)
{
"mcpServers": {
"grafana": {
"command": "docker",
"args": ["run", "--rm", "-i", "--env-file", "/ABSOLUTE/PATH/TO/.env", "grafana-mcp-adapter"]
}
}
}Available Tools
6 toolsgrafana_get_dashboardA
Read-only fetch of a Grafana dashboard's full JSON by uid (panels, variables, queries, meta). Get the uid from grafana_search_dashboards.
Inputs: uid (required).
grafana_healthA
Read-only Grafana health/config check.
Inputs: no named parameters.
grafana_list_datasourcesA
Read-only list of configured Grafana datasources. Returns uid, name, type and is_default. Use a datasource uid with grafana_query.
Inputs: no named parameters.
grafana_logs_linkA
Read-only: build a permanent Grafana Explore (Loki) link for a customer's logs AND return a preview of the most recent lines. Identify the customer/component with free text (e.g. client='april', component='gateway') — it matches case-insensitively against the service_name label, which encodes both. Optionally narrow with line_filter (substring that must appear in the log line). Default range is the last 1 hour; widen with from/to (e.g. from='now-6h'). Returns { query, explore_url, range, preview_count, preview }. Paste explore_url into the ticket; ask the user before widening the range since logs are large.
Inputs: client (required), to, from, limit, +2 more.
grafana_queryA
Read-only metric/log query via Grafana's /api/ds/query. Provide the datasource uid (from grafana_list_datasources), a raw expression (PromQL for Prometheus, LogQL for Loki, etc.), and an optional time range. By default returns a compact per-series digest (labels + count/first/last/min/max/avg); pass raw=true for the full (potentially very large) frames.
Inputs: expr (required), datasource_uid (required), to, raw, +2 more.
grafana_search_dashboardsA
Read-only search of Grafana dashboards (by title substring and/or tags). Returns uid, title, folder, tags and url for each. Use the uid with grafana_get_dashboard to fetch the full dashboard JSON.
Inputs: tags, limit, query.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 6 tools
Each tool targets a distinct Grafana operation: health check, datasource listing, dashboard search and retrieval, general querying, and a specialized logs link builder. No functional overlap exists.
All tools share the 'grafana_' prefix and mostly follow a verb_noun pattern (get_dashboard, list_datasources, search_dashboards). Exceptions like 'health' (noun) and 'logs_link' (noun_noun) are minor inconsistencies within an otherwise coherent scheme.
With 6 tools, the server covers the essential read-only Grafana operations without being excessive. Each tool serves a clear purpose and earns its place.
The set covers core read-only functionality: health, datasources, dashboards, querying, and logs linking. Missing are alerting and annotations, but the surface is complete for common exploration tasks and tools are well-connected (e.g., search_dashboards feeds get_dashboard).
Maintenance
Related MCP Connectors
An MCP server giving access to Grafana dashboards, data and more.
Read-only MCP server for AIStatusDashboard status, incidents, metrics, and fallback recommendations.
Analytics for MCP servers. Query your tool calls, first-call success, retries and schema cost.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables MCP-compatible agents to interact with Grafana instances for searching, creating, and updating dashboards, exploring logs via Loki, querying datasources, managing alerts, incidents, and on-call shifts, and accessing observability data.8Apache 2.0
- FlicenseNot gradedqualityDmaintenanceIn-cluster MCP server for read-only diagnostics of Grafana, Prometheus, Alertmanager, and Loki, enabling metric queries, alert listings, and log queries through Grafana datasource proxies with Kubernetes RBAC authentication.-
- FlicenseNot gradedqualityDmaintenanceMCP server for Grafana that provides tools to interact with dashboards, data sources, alerts, and more.1-
- FlicenseAqualityDmaintenanceRead-only MCP server that exposes Fathom meetings and transcripts as tools for use by skills in the TICKETS workspace.4-