
Grafana MCP, Grafana Labs' official MCP server
Lets your AI assistant search Grafana dashboards, query your metrics and logs, and manage alerts and incidents.
This is Grafana Labs' own server. It lets Claude, Cursor and other assistants search dashboards, query Prometheus and Loki, and manage alerts and incidents. You can run it yourself for free with a service account token, on Grafana Cloud or self-hosted Grafana 9.0 and later. Grafana Cloud also runs a hosted version you sign in to, but it needs Grafana Assistant and bills each user as an Assistant user.
Good for
- On-call engineers who want Claude to dig through metrics, logs and traces when something breaks
- Developers asking why an app is slow, using real CPU, trace and query data
- Teams that want to find dashboards and alert rules by describing them in plain words
- Grafana Cloud teams already paying for Grafana Assistant who prefer a no-install connection
Works in
8 supported · 6 should work
Tap an app to see how to set it up and any known problems.
Documented by Grafana, for both the local and the hosted server. The tool list timed out on older versions.
Known problems
- Fixed
Tool list timed out right after connecting, so no Grafana tools appeared
Workaround: Upgrade to 1.6.1 or later.
Reported Sep 2026 · GitHub
- Fixed
Server showed as connected but fetching tools hung until it timed out
Reported Sep 2026 · GitHub
- Open
The full tool set takes about 16K tokens of context
Workaround: Use --enabled-tools or --disabled-tools to load only the tool groups you need.
Since Feb 2026 · GitHub
Documented by Grafana. The directory extension lags behind releases.
Known problems
- Closed
The extension in Claude's directory lags behind the latest release
Workaround: Download the latest .mcpb from GitHub releases instead of the directory copy.
Reported Sep 2026 · GitHub
- Closed
Panel viewer showed the image but not the Open in Grafana link
Reported Sep 2026 · GitHub
- Fixed
Extension failed to start on Intel Macs
Workaround: Fixed in the 1.5.1 extension; older copies ship an Apple Silicon only binary.
Reported Aug 2026 · GitHub
Documented by Grafana, with a one-click link for the hosted Cloud server.
Known problems
- Closed
Hosted server hid alert rule tools completely when write access was not granted
Reported Sep 2026 · GitHub
Documented by Grafana. Run it locally on port 8000, or add the hosted address. grafana.com
Documented by Grafana, for both the local and the hosted server. grafana.com
Documented by Grafana, using codex mcp add or an entry in config.toml. grafana.com
Documented by Grafana, using an entry in settings.json. grafana.com
Documented by Grafana. In Zed the entry goes under context_servers, not mcpServers. grafana.com
Supports local MCP servers.
Supports hosted MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports hosted MCP servers.
Install mcp-grafana
Claude Code
- 1
Copy this prompt.
Add the mcp-grafana server to Claude Code. Run this command: claude mcp add grafana -e GRAFANA_URL=<GRAFANA_URL> -e GRAFANA_SERVICE_ACCOUNT_TOKEN=<GRAFANA_SERVICE_ACCOUNT_TOKEN> -- uvx mcp-grafana Ask me for GRAFANA_URL, GRAFANA_SERVICE_ACCOUNT_TOKEN and only put the values in that config, nowhere else. Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires uv (Python), Grafana 9.0 or later.
Read-only mode: Add the --disable-write flag to the args (after mcp-grafana) to run in read-only mode. It removes the write tools, including the raw SQL and InfluxDB query tools. For Docker, add it after -t stdio.
Needs Grafana 9.0 or later. Create a service account in Grafana, generate a token, and put it in the env values. Grafana's own tip: if you do not want to set fine-grained permissions, the built-in Editor role covers most tools, but it can write. For Grafana Cloud, use your instance address as the URL. The Docker image defaults to SSE, so the stdio method passes -t stdio and needs -i. GRAFANA_API_KEY is deprecated; use the service account token.
From the project README. Source.
Tools
Finds dashboards by name, folder, tag or star
Finds dashboards by name, folder, tag or star.
Try: “Find my dashboards tagged production”
Read ·
search_dashboardsGives a compact summary of a dashboard
Gives a compact summary of a dashboard.
Try: “Summarize the API latency dashboard”
Read ·
get_dashboard_summaryOpens a full dashboard, or an older saved version
Opens a full dashboard, or an older saved version.
Try: “Open the dashboard with uid abc123”
Read ·
get_dashboard_by_uidLists the queries behind each dashboard panel
Lists the queries behind each dashboard panel.
Try: “What queries power the panels on my latency dashboard?”
Read ·
get_dashboard_panel_queriesPulls one part out of a dashboard
Pulls one part out of a dashboard.
Try: “Get the template variables from this dashboard”
Read ·
get_dashboard_propertyShows the saved versions of a dashboard
Shows the saved versions of a dashboard.
Try: “Show the version history of my latency dashboard”
Read ·
list_dashboard_versionsFinds folders by title
Finds folders by title.
Try: “Find the folder called Platform”
Read ·
search_foldersRuns a Prometheus query
Runs a Prometheus query.
Try: “What is the request rate for the checkout service over the last hour?”
Read ·
query_prometheusSearches your logs with a Loki query
Searches your logs with a Loki query.
Try: “Show error logs from the checkout service in the last 15 minutes”
Read ·
query_loki_logsSearches traces with TraceQL
Searches traces with TraceQL.
Try: “Find traces slower than 2 seconds in the payments service”
Read ·
search_tempo_tracesFetches one trace by ID
Fetches one trace by ID.
Try: “Show me trace 4bf92f3577b34da6”
Read ·
get_tempo_traceLists your connected data sources
Lists your connected data sources.
Try: “What data sources do I have in Grafana?”
Read ·
list_datasourcesShows one data source by UID or name
Shows one data source by UID or name.
Try: “Show the details of my Prometheus data source”
Read ·
get_datasourceChecks whether data sources are working
Checks whether data sources are working.
Try: “Are all my data sources healthy?”
Read ·
check_datasources_healthLists incidents in Grafana Incident
Lists incidents in Grafana Incident.
Try: “List open incidents”
Read ·
list_incidentsShows one incident with its custom fields
Shows one incident with its custom fields.
Try: “Show incident 42”
Read ·
get_incidentLists alert groups from Grafana OnCall
Lists alert groups from Grafana OnCall.
Try: “Show the firing alert groups”
Read ·
list_alert_groupsShows one OnCall alert group
Shows one OnCall alert group.
Try: “Show details for alert group 17”
Read ·
get_alert_groupShows who is on call for a schedule
Shows who is on call for a schedule.
Try: “Who is on call right now for the platform schedule?”
Read ·
get_current_oncall_usersLists OnCall schedules
Lists OnCall schedules.
Try: “List our on-call schedules”
Read ·
list_oncall_schedulesShows details of one on-call shift
Shows details of one on-call shift.
Try: “Show details for that on-call shift”
Read ·
get_oncall_shiftLists OnCall teams
Lists OnCall teams.
Try: “List our OnCall teams”
Read ·
list_oncall_teamsLists OnCall users
Lists OnCall users.
Try: “List the people in OnCall”
Read ·
list_oncall_usersLists and inspects alert rules
Lists and inspects alert rules.
Try: “Show my alert rules for the database”
Read ·
alerting_rules_readLists and inspects alert silences
Lists and inspects alert silences.
Try: “Which alerts are silenced right now?”
Read ·
alerting_silences_readFetches annotations, with filters
Fetches annotations, with filters.
Try: “Show deploy annotations from this week”
Read ·
get_annotationsLists annotation tags
Lists annotation tags.
Try: “What annotation tags exist?”
Read ·
get_annotation_tagsLists dashboard snapshots
Lists dashboard snapshots.
Try: “List my dashboard snapshots”
Read ·
list_snapshotsOpens a dashboard snapshot
Opens a dashboard snapshot.
Try: “Open snapshot with key abc”
Read ·
get_snapshotRenders a dashboard panel or view as a PNG
Renders a dashboard panel or view as a PNG.
Try: “Render the CPU panel as an image”
Read ·
get_panel_imageSearches Grafana documentation
Searches Grafana documentation.
Try: “Search the Grafana docs for Loki label best practices”
Read ·
search_docsFetches a Grafana documentation page
Fetches a Grafana documentation page.
Try: “Open that docs page”
Read ·
get_docSearches the Grafana plugin catalog
Searches the Grafana plugin catalog.
Try: “Is there a plugin for MongoDB?”
Read ·
search_plugin_informationShows an installed plugin
Shows an installed plugin.
Try: “Show the details of the Clock panel plugin”
Read ·
get_pluginShows who you are in Grafana and what you can access
Shows who you are in Grafana and what you can access.
Try: “Who am I logged in as in Grafana?”
Read ·
user_infoLists teams
Lists teams.
Try: “List the teams in Grafana”
Read ·
list_teamsLists users in an organization
Lists users in an organization.
Try: “List the users in my organization”
Read ·
list_users_by_orgLists Grafana roles
Lists Grafana roles.
Try: “List all Grafana roles”
Read ·
list_all_rolesShows details of one role
Shows details of one role.
Try: “Show the permissions of the Editor role”
Read ·
get_role_detailsShows who a role is assigned to
Shows who a role is assigned to.
Try: “Who has the Admin role?”
Read ·
get_role_assignmentsLists roles for users
Lists roles for users.
Try: “What roles does this user have?”
Read ·
list_user_rolesLists roles for teams
Lists roles for teams.
Try: “What roles does the platform team have?”
Read ·
list_team_rolesLists permissions on a resource
Lists permissions on a resource.
Try: “Who can edit this dashboard?”
Read ·
get_resource_permissionsLists available Prometheus metric names
Lists available Prometheus metric names.
Try: “What metrics does Prometheus have?”
Read ·
list_prometheus_metric_namesLists Prometheus metric metadata
Lists Prometheus metric metadata.
Try: “Describe the http_requests_total metric”
Read ·
list_prometheus_metric_metadataLists Prometheus label names
Lists Prometheus label names.
Try: “What labels exist for this selector?”
Read ·
list_prometheus_label_namesLists values of a Prometheus label
Lists values of a Prometheus label.
Try: “What values does the job label have?”
Read ·
list_prometheus_label_valuesCalculates percentiles from a Prometheus histogram
Calculates percentiles from a Prometheus histogram.
Try: “What is the p95 request latency?”
Read ·
query_prometheus_histogramLists log label names
Lists log label names.
Try: “What log labels are available?”
Read ·
list_loki_label_namesLists values of a log label
Lists values of a log label.
Try: “What services appear in the logs?”
Read ·
list_loki_label_valuesGets statistics about log streams
Gets statistics about log streams.
Try: “How much log data does the checkout service produce?”
Read ·
query_loki_statsFinds common patterns in logs
Finds common patterns in logs.
Try: “What are the common log patterns for the API?”
Read ·
query_loki_patternsAudits your Loki label setup
Audits your Loki label setup.
Try: “Audit my Loki labels”
Read ·
analyze_loki_labelsLists trace attribute names
Lists trace attribute names.
Try: “What span attributes can I search on?”
Read ·
list_tempo_attribute_namesLists values of a trace attribute
Lists values of a trace attribute.
Try: “What values does service.name have?”
Read ·
list_tempo_attribute_valuesComputes metrics from traces
Computes metrics from traces.
Try: “Chart the error rate from traces”
Read ·
query_tempo_metricsCompares two traces
Compares two traces.
Try: “Compare these two traces”
Read ·
diff_tempo_tracesLists Pyroscope label names
Lists Pyroscope label names.
Try: “What Pyroscope labels exist?”
Read ·
list_pyroscope_label_namesLists values of a Pyroscope label
Lists values of a Pyroscope label.
Try: “What services are profiled?”
Read ·
list_pyroscope_label_valuesLists available profile types
Lists available profile types.
Try: “What profile types are available?”
Read ·
list_pyroscope_profile_typesQueries profiles and metrics from Pyroscope
Queries profiles and metrics from Pyroscope.
Try: “Show CPU profile for the API service”
Read ·
query_pyroscopeGets the assertion summary for an entity
Gets the assertion summary for an entity.
Try: “Show assertions for the checkout service”
Read ·
get_assertionsLists the custom fields set up for incidents
Lists the custom fields set up for incidents.
Try: “What custom fields do incidents have?”
Read ·
list_incident_custom_fieldsLists provisioning repositories and their sync state
Lists provisioning repositories and their sync state.
Try: “Show my git-sync repositories”
Read ·
list_provisioning_repositoriesDry-run checks a provisioning file
Dry-run checks a provisioning file.
Try: “Validate this provisioning file”
Read ·
validate_provisioning_fileAsks Grafana Assistant a question
Asks Grafana Assistant a question. It sends your prompt to Grafana Assistant and returns its full text reply.
Try: “Ask Grafana Assistant why latency spiked”
Read ·
ask_assistantOpens a new incident
Opens a new incident.
Try: “Create a high-severity incident for the checkout outage”
Write ·
create_incidentUpdates an incident's status, severity, title or fields
Updates an incident's status, severity, title or fields.
Try: “Mark incident 42 as resolved”
Write ·
update_incidentAdds a note to an incident
Adds a note to an incident.
Try: “Add a note to incident 42 saying we rolled back”
Write ·
add_activity_to_incidentAcknowledges or resolves an OnCall alert group
Acknowledges or resolves an OnCall alert group.
Try: “Acknowledge alert group 17”
Write ·
update_alert_groupUpdates a dashboard or creates a new one
Updates a dashboard or creates a new one.
Try: “Add a CPU panel to my service dashboard”
Write ·
update_dashboardCreates a folder
Creates a folder.
Try: “Create a folder called Platform”
Write ·
create_folderCreates a data source
Creates a data source.
Try: “Add a new Prometheus data source”
Write ·
create_datasourceUpdates a data source
Updates a data source.
Try: “Change the URL of my Prometheus data source”
Write ·
update_datasourceAdds an annotation
Adds an annotation.
Try: “Annotate the dashboard with today's deploy”
Write ·
create_annotationEdits an annotation
Edits an annotation.
Try: “Fix the text of that annotation”
Write ·
update_annotationDeletes an annotation
Deletes an annotation.
Try: “Delete annotation 12”
Delete ·
delete_annotationCreates a dashboard snapshot
Creates a dashboard snapshot.
Try: “Make a snapshot of my latency dashboard”
Write ·
create_snapshotDeletes a dashboard snapshot
Deletes a dashboard snapshot.
Try: “Delete that snapshot”
Delete ·
delete_snapshotCreates, edits and deletes alert rules
Creates, edits and deletes alert rules.
Try: “Create an alert for error rate above 5%”
Delete ·
alerting_rules_writeManages notification policies, contact points and time intervals
Manages notification policies, contact points and time intervals.
Try: “Show how alerts are routed to Slack”
Delete ·
alerting_manage_routingCreates contact points
Creates contact points.
Try: “Add a Slack contact point”
Write ·
alerting_routing_writeCreates, edits and expires alert silences
Creates, edits and expires alert silences.
Try: “Silence the disk alert for 2 hours”
Write ·
alerting_silences_writeInstalls a plugin
Installs a plugin.
Try: “Install the Clock panel plugin”
Write ·
install_pluginWrites an Alloy config snippet for approved log labels
Writes an Alloy config snippet for approved log labels.
Try: “Write an Alloy snippet that keeps only approved labels”
Read ·
suggest_loki_alloy_label_configCalls any Grafana HTTP API endpoint
Calls any Grafana HTTP API endpoint.
Try: “Call the Grafana health API”
Delete ·
grafana_api_requestRuns the query behind a dashboard panel
Runs the query behind a dashboard panel. The README says these tools are off by default; add runpanelquery to the enabled-tools setting to use them.
Try: “Run the query on my CPU panel for the last 6 hours”
Read ·
run_panel_queryGives example queries for a data source type
Gives example queries for a data source type. The README says these tools are off by default; add examples to the enabled-tools setting to use them.
Try: “Show example Prometheus queries”
Read ·
get_query_examplesQueries InfluxDB
Queries InfluxDB. The README says these tools are off by default; add influxdb to the enabled-tools setting to use them.
Try: “Query InfluxDB for CPU usage”
Write ·
query_influxdbLists databases from a SQL data source
Lists databases from a SQL data source. The README says these tools are off by default; add sql to the enabled-tools setting to use them.
Try: “What databases are in my SQL data source?”
Read ·
list_sql_databasesLists tables in a SQL data source
Lists tables in a SQL data source. The README says these tools are off by default; add sql to the enabled-tools setting to use them.
Try: “List the tables”
Read ·
list_sql_tablesShows a table's columns
Shows a table's columns. The README says these tools are off by default; add sql to the enabled-tools setting to use them.
Try: “Describe the orders table”
Read ·
describe_sql_tableRuns a SQL query
Runs a SQL query. The README says these tools are off by default; add sql to the enabled-tools setting to use them.
Try: “Count orders from yesterday”
Delete ·
query_sqlLists CloudWatch namespaces
Lists CloudWatch namespaces. The README says these tools are off by default; add cloudwatch to the enabled-tools setting to use them.
Try: “List my CloudWatch namespaces”
Read ·
list_cloudwatch_namespacesLists metrics in a CloudWatch namespace
Lists metrics in a CloudWatch namespace. The README says these tools are off by default; add cloudwatch to the enabled-tools setting to use them.
Try: “What metrics are in AWS/EC2?”
Read ·
list_cloudwatch_metricsLists dimensions for a CloudWatch metric
Lists dimensions for a CloudWatch metric. The README says these tools are off by default; add cloudwatch to the enabled-tools setting to use them.
Try: “What dimensions does CPUUtilization have?”
Read ·
list_cloudwatch_dimensionsLists values for a CloudWatch dimension
Lists values for a CloudWatch dimension. The README says these tools are off by default; add cloudwatch to the enabled-tools setting to use them.
Try: “List instance IDs”
Read ·
list_cloudwatch_dimension_valuesQueries CloudWatch metrics
Queries CloudWatch metrics. The README says these tools are off by default; add cloudwatch to the enabled-tools setting to use them.
Try: “Show EC2 CPU for the last hour”
Read ·
query_cloudwatchLists CloudWatch log groups
Lists CloudWatch log groups. The README says these tools are off by default; add cloudwatch to the enabled-tools setting to use them.
Try: “List my log groups”
Read ·
list_cloudwatch_log_groupsLists fields in a CloudWatch log group
Lists fields in a CloudWatch log group. The README says these tools are off by default; add cloudwatch to the enabled-tools setting to use them.
Try: “What fields does this log group have?”
Read ·
list_cloudwatch_log_group_fieldsRuns a CloudWatch Logs Insights query
Runs a CloudWatch Logs Insights query. The README says these tools are off by default; add cloudwatch to the enabled-tools setting to use them.
Try: “Find errors in the API log group”
Read ·
query_cloudwatch_logsLists Google Cloud projects you can read logs from
Lists Google Cloud projects you can read logs from. The README says these tools are off by default; add cloudlogging to the enabled-tools setting to use them.
Try: “Which GCP projects can I query?”
Read ·
list_cloud_logging_projectsLists log buckets in a Google Cloud project
Lists log buckets in a Google Cloud project. The README says these tools are off by default; add cloudlogging to the enabled-tools setting to use them.
Try: “List log buckets”
Read ·
list_cloud_logging_bucketsLists log views in a bucket
Lists log views in a bucket. The README says these tools are off by default; add cloudlogging to the enabled-tools setting to use them.
Try: “List log views”
Read ·
list_cloud_logging_viewsQueries Google Cloud Logging
Queries Google Cloud Logging. The README says these tools are off by default; add cloudlogging to the enabled-tools setting to use them.
Try: “Show errors from Cloud Run”
Read ·
query_cloud_loggingQueries Elasticsearch or OpenSearch
Queries Elasticsearch or OpenSearch. The README says these tools are off by default; add elasticsearch to the enabled-tools setting to use them.
Try: “Search for failed logins”
Read ·
query_elasticsearchQueries Quickwit
Queries Quickwit. The README says these tools are off by default; add quickwit to the enabled-tools setting to use them.
Try: “Search Quickwit for timeouts”
Read ·
query_quickwitQueries Graphite metrics
Queries Graphite metrics. The README says these tools are off by default; add graphite to the enabled-tools setting to use them.
Try: “Query Graphite for request counts”
Read ·
query_graphiteQueries Graphite metric density
Queries Graphite metric density. The README says these tools are off by default; add graphite to the enabled-tools setting to use them.
Try: “Show Graphite metric density”
Read ·
query_graphite_densityLists Graphite metrics
Lists Graphite metrics. The README says these tools are off by default; add graphite to the enabled-tools setting to use them.
Try: “List Graphite metrics”
Read ·
list_graphite_metricsLists Graphite tags
Lists Graphite tags. The README says these tools are off by default; add graphite to the enabled-tools setting to use them.
Try: “List Graphite tags”
Read ·
list_graphite_tagsLists, searches and fetches LLM conversations
Lists, searches and fetches LLM conversations.
Try: “Find recent conversations for my support agent”
Read ·
agento11y_manage_conversationsFetches LLM generation details and evaluation scores
Fetches LLM generation details and evaluation scores.
Try: “Show scores for that generation”
Read ·
agento11y_manage_generationsReads the agent catalog and version history
Reads the agent catalog and version history.
Try: “List my agents”
Read ·
agento11y_manage_agentsManages evaluators, templates and the judge catalog
Manages evaluators, templates and the judge catalog.
Try: “List my evaluators”
Delete ·
agento11y_manage_evaluatorsManages eval rules and guards
Manages eval rules and guards.
Try: “List my eval rules”
Delete ·
agento11y_manage_eval_rulesManages saved conversations and their collections
Manages saved conversations and their collections.
Try: “Save this conversation to a collection”
Delete ·
agento11y_manage_eval_collectionsReads offline experiments and their results
Reads offline experiments and their results.
Try: “Show results for my latest experiment”
Write ·
agento11y_manage_experimentsManages test suites and test cases for experiments
Manages test suites and test cases for experiments.
Try: “List my test suites”
Delete ·
agento11y_manage_test_suitesUsed by the assistant behind the scenes
Describes a type of Grafana resource
Describes a type of Grafana resource.
Try: “Describe the dashboard resource type”
Read ·
get_resource_descriptionGets TraceQL documentation
Gets TraceQL documentation.
Try: “Show me how TraceQL works”
Read ·
get_tempo_traceql_docsBuilds a link to a Grafana resource
Builds a link to a Grafana resource.
Try: “Give me a link to that dashboard”
Read ·
generate_deeplink
Read from the source code
Pros and cons
Good
- Made and maintained by Grafana Labs.
- Free and open source under Apache-2.0 for Grafana Cloud or your own Grafana.
- Covers dashboards, Prometheus and Loki queries, alerting, incidents and OnCall in one server.
- Read-only mode with one flag, plus a Viewer service account for tighter access.
- Load only the tool groups you need to save context.
- Hosted Grafana Cloud option needs no install and signs you in with your own account.
Watch out
- The full tool set uses about 16K tokens of context.
- Write tools are on by default.
- Raw SQL tools can change data unless read-only mode is on.
- Needs Grafana 9.0 or later.
- The hosted server is Grafana Cloud only, needs Grafana Assistant and bills each user.
- Cloud access policy tokens do not work, only service account tokens.
Permissions
medium risk · 6 things to watch
Medium risk
The assistant can change dashboards and alert rules, and run SQL that edits or drops data, if the token you give it allows that. Start in read-only mode and use a Viewer service account.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only modeHow to turn it on
Watch out for
Write tools are on by default
Unless you turn on read-only mode, the assistant can create and update dashboards, alert rules, incidents and annotations, and delete alert rules. Start with read-only and add write access later. github.com
SQL tools run any statement the database allows
A DROP TABLE or DELETE runs if the datasource login permits it. Read-only mode removes these tools; on the hosted server, clear Query as well as Write. github.com
Network modes need a caller token
The Docker image listens on all network addresses in SSE or streamable-http mode. Set a server token so callers must send it. github.com
The Editor role gives more access than needed
Grafana's docs offer Editor as the easy choice for the service account. Use Viewer, or the per-tool permissions in the README, unless you need write tools. github.com
Usage stats go to Grafana Labs by default
Tool names, failure counts and flag names are sent; queries and credentials are not. Set DO_NOT_TRACK=1 to stop it. github.com
The hosted server acts as you
It uses your own Grafana permissions. You can revoke a connection under Settings, Connectors, MCP clients. grafana.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Turning on read-only | Start the server with --disable-write: write tools and raw SQL/InfluxDB query tools are removed, PromQL/LogQL queries stay. Pair it with a Viewer-level service account. On the hosted Cloud server, clear Write and Query access on the consent page. |
|---|---|
| Talks to | Your Grafana instance. Grafana Labs (anonymous usage stats, can be turned off) |
Alternatives
Use this server if your metrics and logs are in Grafana. Pick the Prometheus or VictoriaMetrics server if you query those directly, SigNoz if you run it instead of Grafana, and mcp-k6 for load tests.
| Server | Official | Runs | Can change things | GitHub stars |
|---|---|---|---|---|
| For Grafana | ||||
| GrafanaThis page | Official | HostedLocal | Yes | 3.6K |
| mcp-k6 | Official | – | – | 56 |
| Other tools for the same job | ||||
| Prometheus MCP Server | Community | Local | – | 521 |
| mcp-victoriametrics | Community | Local | – | 240 |
| SigNoz MCP ServerFor SigNoz | Official | HostedLocal | – | 130 |
What people say
People like how much debugging time it saves and dislike how much context its full tool set uses.
What people like
- Engineers say it saves days of debugging when paired with read-only kubectl in Claude Code.
- An agent can pull metrics, traces and CPU profiles itself and work out why an app is slow.
- It works against a small self-hosted Grafana as well as Grafana Cloud.
Common complaints
- The default tool set is about 16K tokens, close to 10% of Claude's context.
- The Claude Desktop directory copy trails releases, so some users install extra tools to get a working version.
- Only service account tokens or basic sign-in work, not Grafana Cloud access policy tokens.
In their words
“I have to say though, giving Amp/Claude Code the Grafana MCP + read-only kubectl has saved me days worth of debugging.”bittermandelHacker NewsAug 2026
“In my case LLM goes to grafana mcp, pulls metrics/traces/cpu profiles. Figures out what is slow and proposes a solution.”0x457GitHub issueFeb 2026
“Currently the MCP is ~16K tokens. That's just short of 10% of available context for Claude.”JohnieLeeGitHub issueSep 2026
“Non-IT guys have to install brew, uv and so on to use latest version that works properly”playments-admin
Questions
Is the Grafana MCP server free?
The server you run yourself is free and open source. The hosted server at mcp.grafana.com is billed as part of Grafana Assistant, and each user who connects counts as an active Assistant user. Read-only use does not use Assistant tokens.
Should I use the hosted server or run it myself?
Run it yourself if you use self-hosted Grafana, want it free, or do not have Grafana Assistant. Use the hosted server if you are on Grafana Cloud, have Assistant enabled, and want sign-in with no token or install.
What is the Grafana MCP server URL?
The hosted Grafana Cloud address is https://mcp.grafana.com/mcp. For several Cloud stacks, add one entry per stack ending in /mcp/<your-stack>.grafana.net. The self-run version has no fixed address.
Which Grafana versions does it need?
Grafana 9.0 or later. On older versions the datasource tools fail with an "id is invalid" error, and the maintainers do not plan to support 8.5. A few tools need more, such as the Google Cloud Logging tools, which need Grafana 11.2.
How do I make it read-only?
Start the server with the --disable-write option. That removes create, update and delete tools and the raw SQL tools, while metric and log queries keep working. On the hosted server, clear Write and Query on the sign-in page; clearing only Write still allows raw SQL.
What permissions does the service account need?
Enough for the tools you use; the README lists the exact permissions per tool. Use a Viewer account for read-only work. The built-in Editor role is simpler but gives broad write access.
How much of my context window does it use?
The full tool set takes about 16K tokens, close to 10% of Claude's context. You can load only the tool groups you need to cut that down.
Can it delete things?
Yes, unless read-only mode is on. It can delete alert rules, and its SQL tools run any statement your datasource login allows.
Does it send data to Grafana Labs?
By default it sends anonymous usage stats: which tools ran, how many failed and flag names. Queries, log lines and credentials are not sent, and you can turn it off.
Why does Docker connect but my client sees no server?
The Docker image starts in a network mode by default. For Claude Desktop, Cursor and similar apps, add the stdio options to the docker run arguments, as the README shows.
Examples
Find out why an app is slow from metrics, traces and CPU profiles
Why is the checkout service slow since yesterday? Check metrics, traces and CPU profiles.
Explain a resource trend on a self-hosted Grafana
Why has CPU been increasing this week?
SourceDebug an infrastructure problem from Claude Code alongside read-only kubectl
SourceCheck observability data while QA-testing a web server against a local Grafana
Source
Information
updated Nov 2, 2026
- Publisher
- Grafana Labs (official)
- Package
grafana/mcp-grafana- Registry
- Docker
- Endpoint
https://mcp.grafana.com/mcp- Transport
- stdio, SSE, Streamable HTTP
- Version
- v2.1.1
- License
- Apache-2.0
- Last commit
- Oct 31, 2026
- Open issues
- 56
- Listed in
- Official MCP registry, Docker MCP Catalog
Grafana MCP server docsGitHubDockerOfficial MCP RegistrySet up by clientChangelogIssuesGrafana Cloud hosted MCP server
Other Grafana MCP servers
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 18Hide
- mcp-grafana README · github.com
- grafana.com · grafana.com
- grafana.com · grafana.com
- github.com · github.com
- github.com · github.com
- Grafana MCP client setup guides · grafana.com
- github.com · github.com
- github.com · github.com
- github.com · github.com
- github.com · github.com
- github.com · github.com
- github.com · github.com
- news.ycombinator.com · news.ycombinator.com
- news.ycombinator.com · news.ycombinator.com
- github.com · github.com
- news.ycombinator.com · news.ycombinator.com
- news.ycombinator.com · news.ycombinator.com
- SigNoz MCP Server · signoz.io