mshegolev/prometheus-mcp
This server provides read-only access to a Prometheus monitoring instance, enabling AI agents and MCP-capable clients to query metrics, investigate alerts, and explore scrape targets.
List Metrics (
prometheus_list_metrics): Discover all metric names with optional case-insensitive substring filtering (e.g., metrics containing 'http'). Results are capped at 500 metrics.Instant PromQL Query (
prometheus_query): Execute any PromQL expression at a specific point in time (or now), returning result type, sample count, labels, timestamps, and values — ideal for checking current state.Range PromQL Query (
prometheus_query_range): Execute a PromQL expression over a time window with a specified resolution step, returning time-series data — useful for graphing trends or investigating past incidents.List Active Alerts (
prometheus_list_alerts): Retrieve all firing or pending alerts, including labels (e.g.,alertname,severity), state, activation time, and current value, with a summary grouped by state and severity.List Scrape Targets (
prometheus_list_targets): Inspect all scrape targets with job name, instance address, health status (up/down/unknown), last scrape duration, and any errors — filterable by state (active,dropped, orany).
Key characteristics: All tools are strictly read-only (no risk of modifying Prometheus data), support Bearer token or HTTP Basic authentication, return structured JSON output, and are configurable via environment variables.
Provides tools for querying Prometheus metrics with PromQL, listing available metric names, inspecting active alerts, and exploring scrape targets.
prometheus-mcp
MCP server for Prometheus metrics and observability. Give Claude (or any MCP-capable agent) read access to your Prometheus instance — query metrics with PromQL, inspect active alerts, and explore scrape targets — without leaving the conversation.
Why another Prometheus MCP?
The existing Prometheus integrations require custom scripts or direct API knowledge. This server:
Speaks the standard Model Context Protocol over stdio — works with Claude Desktop, Claude Code, Cursor, and any MCP client.
Is read-only: all 5 tools carry
readOnlyHint: true— zero risk of modifying Prometheus data.Returns dual-channel output: structured JSON (
structuredContent) for programmatic use + Markdown (content) for human-readable display.Has actionable error messages that name the exact env var to fix and suggest a next step.
Supports Bearer token, HTTP Basic auth, or no auth (common for internal deployments).
Related MCP server: Proms MCP Server
Tools
Tool | Endpoint | Description |
|
| List all metric names with optional substring filter (cap 500) |
|
| Execute an instant PromQL query |
|
| Execute a PromQL range query returning time-series |
|
| List active and pending alerts |
|
| List scrape targets by health and job |
v4.0 Advanced Alert Correlation Features
Version 4.0 introduces powerful new capabilities for AI agents to autonomously investigate production errors:
Cross-Instance Alert Correlation
Automatically identify related alerts across multiple Prometheus instances
Group alerts by service identifiers to understand incident scope
Detect cascading alert patterns with directional dependency inference
Root Cause Analysis
Anomaly detection in metrics with automatic seasonality adjustment
Dependency chain traversal from symptoms to potential root causes
Change point detection correlating alerts with recent deployments or config changes
Ranked root cause candidates based on evidence strength and impact analysis
Dependency Mapping & Health
Dynamic service dependency maps built from traffic correlation analysis
Cross-cluster dependency visualization showing service interoperation
Synthetic health probing to assess dependency resilience
Load shedding recommendations based on dependency fragility
Trend Analysis & Benchmarking
Historical pattern recognition for recurring alert schedules
Capacity forecasting to predict resource exhaustion
MTTR benchmarking comparing resolution times against historical data
Deviation detection triggering higher-priority notifications for pattern breaks
Integrated Analysis Tool
New
federation_analyze_alertstool combining all v4.0 featuresUnified output format optimized for AI agent consumption
Comprehensive incident context in a single tool call
Installation
pip install prometheus-mcpOr run directly without installing:
uvx prometheus-mcpConfiguration
All configuration is via environment variables:
Variable | Required | Default | Description |
| Yes | — | Prometheus server URL, e.g. |
| No | — | Bearer token (takes precedence over Basic auth) |
| No | — | HTTP Basic auth username |
| No | — | HTTP Basic auth password |
| No |
| Set |
Copy .env.example to .env and fill in your values.
Claude Desktop / Claude Code setup
Add to your MCP config (claude_desktop_config.json or .claude/mcp.json):
{
"mcpServers": {
"prometheus": {
"command": "prometheus-mcp",
"env": {
"PROMETHEUS_URL": "https://prometheus.example.com",
"PROMETHEUS_TOKEN": "your-token-here"
}
}
}
}Or with uvx (no install required):
{
"mcpServers": {
"prometheus": {
"command": "uvx",
"args": ["prometheus-mcp"],
"env": {
"PROMETHEUS_URL": "https://prometheus.example.com"
}
}
}
}Docker
docker run --rm -e PROMETHEUS_URL=https://prometheus.example.com prometheus-mcpExample queries
Once configured, ask Claude:
"What metrics does Prometheus have about HTTP requests?"
"What is the current request rate for the payment service?"
"Show me CPU usage over the last hour with 5-minute resolution"
"Are there any firing alerts? What's their severity?"
"Which scrape targets are currently down and why?"
"How many node-exporter instances are up?"
Tool usage guide
prometheus_list_metrics
Returns all metric names Prometheus knows about. Use pattern to filter by substring (case-insensitive). Start here when you don't know which metrics are available. Output is capped at 500 metrics with a truncation hint.
prometheus_query
Execute an instant PromQL expression and get current values. Returns result type (vector/scalar/matrix/string), sample count, and per-sample labels and values.
Parameters:
query(required) — PromQL expression, e.g.up,rate(http_requests_total[5m])time(optional) — RFC3339 or Unix timestamp; defaults to now
prometheus_query_range
Execute a PromQL expression over a time window. Returns one series per matching time series with timestamped values. Total data points across all series are capped at 5000.
Parameters:
query(required) — PromQL expressionstart/end(required) — RFC3339 or Unix timestampsstep(required) — resolution like15s,1m,5m
Prometheus rejects steps that would produce > 11,000 points per series (HTTP 422). Increase step or narrow the range if this happens.
Note: The Prometheus range API does not support filtering by branch or commit — filters are expressed purely in PromQL label matchers.
prometheus_list_alerts
Returns all active/pending alerts with labels (including alertname, severity), state, activation time, and current value. Includes a state summary (firing vs pending counts).
prometheus_list_targets
Returns scrape targets with job name, instance address, health (up/down/unknown), last scrape duration in milliseconds, and any error message. Includes a per-job summary. Filter by state: active (default), dropped, or any.
Performance characteristics
All tools use a single persistent
requests.Sessionwith connection pooling.The session has
trust_env = Falseto bypass environment proxies (Prometheus is typically an internal service).Requests time out after 30 seconds.
prometheus_query_rangecaps output at 5000 total points across all series — use a larger step for long windows.prometheus_list_metricsreturns up to 500 metrics after filtering.
Development
git clone https://github.com/mshegolev/prometheus-mcp
cd prometheus-mcp
pip install -e '.[dev]'
pytest tests/ -v
ruff check src tests
ruff format src testsAPI Specification
This project includes an OpenAPI 3.0 specification in the specs/ directory that documents all MCP tools exposed by the server.
To validate the specification:
python3 specs/validate_spec.pyAutomation
This repository includes automated scripts and GitHub Actions workflows to streamline the release process:
Scripts
scripts/auto-commit-push.sh- Automatically commit and push changes with optional release triggerscripts/release.sh- Full release automation including pipeline checking, version bumping, and tagging
GitHub Actions Workflows
post-push-check.yml- Monitors test pipeline status after each push and comments on the commitauto-release.yml- Manual workflow to create releases with version bumping (patch, minor, or major)
To trigger an automated release:
Go to the Actions tab in GitHub
Select "Auto Release" workflow
Run the workflow with your preferred version bump type
License
MIT — see LICENSE.
Available Tools
21 toolsalertmanager_get_statusARead-onlyIdempotent
Get Alertmanager cluster status, version, and config.
Wraps GET /api/v2/status. Returns cluster state (ready/settling),
version info, uptime, and the raw configuration YAML.
Examples:
- Use when: "Is Alertmanager healthy?"
→ check cluster_status.
- Use when: "What version of Alertmanager is running?"
→ check version_info.
- Don't use when: You want to see active alerts
(call alertmanager_list_alerts).
Returns:
dict with cluster_status / version_info / uptime /
config_yaml.
Inputs: instance. Returns: uptime, config_yaml, version_info, cluster_status
alertmanager_list_alert_groupsARead-onlyIdempotent
List alert groups from Alertmanager showing routing topology.
Wraps GET /api/v2/alerts/groups. Returns groups with their labels,
receiver, and alert count — shows how alerts are grouped for notification.
Examples:
- Use when: "Why did I get one notification instead of many?"
→ check which alerts are in the same group.
- Use when: "What receiver handles payment alerts?"
→ find the group and check its receiver.
- Don't use when: You want individual alert details
(call alertmanager_list_alerts).
Returns:
dict with total_groups / total_alerts / groups
(list with labels, receiver, alert_count).
Inputs: instance. Returns: groups, total_alerts, total_groups
alertmanager_list_alertsARead-onlyIdempotent
List alerts from Alertmanager with suppression state.
Wraps GET /api/v2/alerts. Returns alerts with their status
(active/suppressed/unprocessed), silence IDs, and inhibition IDs.
Unlike prometheus_list_alerts, this shows WHY an alert is or
isn't firing — suppressed alerts include silencedBy and
inhibitedBy arrays.
Examples:
- Use when: "Why isn't the HighCPU alert firing?"
→ check if it's suppressed (silencedBy or inhibitedBy).
- Use when: "Show all suppressed alerts"
→ filter by status.state == 'suppressed'.
- Don't use when: You want Prometheus-side alert state
(call prometheus_list_alerts).
Returns:
dict with total_count / active_count / suppressed_count /
unprocessed_count / alerts (list with status, silencedBy, etc.).
Inputs: instance. Returns: alerts, total_count, active_count, suppressed_count, +1 more
alertmanager_list_silencesARead-onlyIdempotent
List all silences from Alertmanager.
Wraps GET /api/v2/silences. Returns silences with matchers, status
(active/pending/expired), creator, comment, and time bounds.
Use this to understand which alerts are currently silenced and why.
During incident handoffs, knowing what's silenced is critical — a
silenced alert is invisible in Prometheus /alerts.
Examples:
- Use when: "Is the HighCPU alert silenced?"
→ search silences for matching matchers.
- Use when: "Who silenced alerts for the payment service?"
→ check createdBy and comment.
- Don't use when: You want active firing alerts
(call alertmanager_list_alerts).
Returns:
dict with total_count / active_count / pending_count /
expired_count / silences (list with matchers, status, etc.).
Inputs: instance, instances. Returns: silences, total_count, active_count, expired_count, +1 more
correlate_alerts_across_instancesARead-onlyIdempotent
Correlate alerts across multiple Prometheus instances.
Identifies related alerts that fire simultaneously or in sequence across different Prometheus instances using temporal windows and label similarity.
Use this to:
Understand cross-instance incident scope
Identify related alerts in different clusters/regions
Detect systemic issues affecting multiple instances
Examples: - "Are there related alerts firing across our US and EU clusters?" - "Show me alerts that might be related to this HighCPU alert"
Returns: CorrelationResult with correlated alerts, groups, and cascades.
Inputs: instance, instances, enable_rca, temporal_window, +1 more. Returns: groups, cascades, rca_enhancement, correlated_alerts, +2 more
detect_cascading_alertsARead-onlyIdempotent
Detect cascading alert patterns with directional dependency inference.
Identifies alert propagation patterns that indicate dependency failures.
Use this to:
Trace failure propagation paths through your system
Identify root cause candidates for complex incidents
Understand service dependency relationships
Examples: - "What alerts typically fire after DatabaseConnectionFailed?" - "Show me the failure propagation chain in this incident"
Returns: CascadeDetectionResult with detected cascades and root causes.
Inputs: instance, instances, temporal_window. Returns: cascades, root_causes, total_cascades, rca_enhancement
federation_list_instancesARead-onlyIdempotent
List all configured Prometheus and Alertmanager instances with health status
Inputs: no named parameters. Returns: instances, total_count, federation_enabled
group_alerts_by_serviceARead-onlyIdempotent
Group alerts by service identifiers across all instances.
Clusters related alerts into service-level incident analysis bundles.
Use this to:
Understand which services are affected by current incidents
Focus incident response efforts on specific service teams
Identify services with multiple simultaneous alerts
Examples: - "Which services are currently experiencing alerts?" - "Group all alerts by service for my incident report"
Returns: AlertGroupResult with alerts grouped by service identifier.
Inputs: instance, instances. Returns: groups, total_groups, rca_enhancement, ungrouped_count
prometheus_get_build_infoARead-onlyIdempotent
Get Prometheus build information.
Wraps GET /api/v1/status/buildinfo. Returns version, Go version,
Git revision, branch, build user, and build date.
Examples:
- Use when: "What version of Prometheus is running?" → check version.
- Use when: Debugging version-specific behavior.
- Don't use when: You want runtime stats
(call prometheus_get_runtime_info).
Returns:
dict with version / revision / branch / buildUser /
buildDate / goVersion.
Inputs: instance. Returns: branch, version, revision, buildDate, +2 more
prometheus_get_cardinalityARead-onlyIdempotent
Get TSDB statistics and cardinality data from Prometheus.
Wraps GET /api/v1/status/tsdb. Returns head stats (total series,
chunks, time range) and top-N lists: metrics by series count, labels
by value count, and labels by memory usage.
Use this to investigate cardinality explosions — the #1 operational Prometheus problem. High series counts slow queries and increase memory.
Examples:
- Use when: "Why is Prometheus using so much memory?"
→ check num_series and top_metrics_by_series.
- Use when: "Which labels have the most values?"
→ check top_labels_by_value_count.
- Don't use when: You want current metric values
(call prometheus_query).
Returns:
dict with num_series / chunk_count / min_time / max_time /
top_metrics_by_series / top_labels_by_value_count /
top_labels_by_memory_bytes.
Inputs: instance. Returns: max_time, min_time, num_series, chunk_count, +4 more
prometheus_get_metric_metadataARead-onlyIdempotent
Get metric metadata (HELP text, TYPE, UNIT) from Prometheus.
Wraps GET /api/v1/metadata. Returns the metadata that Prometheus
scraped from HELP, TYPE, and UNIT lines in the exposition
format. Each metric may have multiple metadata entries if different
scrape targets expose different help strings.
Use this to understand what a metric measures, its type (counter, gauge,
histogram, summary), and unit — essential for writing correct PromQL.
For example, knowing a metric is a counter means you should use rate()
or increase(); a gauge can be used directly.
Examples:
- Use when: "What does http_requests_total measure?"
→ metric='http_requests_total'; read help and type.
- Use when: "Show me all histogram metrics"
→ call with no filter; filter results where type='histogram'.
- Use when: Starting an investigation — check metric types before
writing PromQL to avoid using rate() on a gauge.
- Don't use when: You already know the metric type and want to
query values (call prometheus_query directly).
Returns:
dict with metric / total_count / returned_count /
truncated / metadata (dict of metric name → list of
{type, help, unit}).
Inputs: metric, instance. Returns: metric, metadata, truncated, total_count, +1 more
prometheus_get_runtime_infoARead-onlyIdempotent
Get Prometheus runtime information.
Wraps GET /api/v1/status/runtimeinfo. Returns operational data:
goroutine count, time series count, storage retention policy,
start time, corruption count, and config reload status.
Examples:
- Use when: "Why is Prometheus slow?" → check goroutine count
and time series count.
- Use when: "What's the retention policy?" → check storage_retention.
- Don't use when: You want the Prometheus version
(call prometheus_get_build_info).
Returns:
dict with start_time / goroutine_count / time_series_count /
storage_retention / corruptionCount / reloadConfigSuccess /
lastConfigTime.
Inputs: instance. Returns: start_time, lastConfigTime, corruptionCount, goroutine_count, +3 more
prometheus_health_checkARead-onlyIdempotent
Check Prometheus liveness and readiness.
Calls GET /-/healthy and GET /-/ready — management endpoints
outside the /api/v1 namespace. Returns whether each probe returned
a 200 status code.
Use this to verify Prometheus is actually running before investigating blank query results. A failed health check means Prometheus is down; a failed readiness check means it's starting up or shutting down.
Examples:
- Use when: "Why are all my queries returning empty results?"
→ check if Prometheus is healthy first.
- Use when: Setting up a new MCP connection — verify the target
is reachable and healthy.
- Don't use when: You want metric values (call prometheus_query).
Returns:
dict with healthy (bool), healthy_status_code,
ready (bool), ready_status_code.
Inputs: instance. Returns: ready, healthy, ready_status_code, healthy_status_code
prometheus_list_alertsARead-onlyIdempotent
List all active and pending alerts from Prometheus.
Wraps GET /api/v1/alerts. Returns every alert that Prometheus currently
tracks, with labels (including alertname, severity), state
(firing / pending), the time it became active, and its current value.
Also returns a summary grouped by state and a count by severity label.
Examples:
- Use when: "Are there any firing alerts right now?"
→ check firing_count and alerts where state='firing'.
- Use when: "Show me all critical alerts"
→ filter alerts by labels.severity == 'critical'.
- Use when: Checking system health during an incident — list alerts
first to understand what's firing before querying metrics.
- Don't use when: You want historical alert data (Prometheus stores only
current state; use Alertmanager or a recording rule for history).
- Don't use when: You want raw metric values
(call prometheus_query or prometheus_query_range).
Returns:
dict with total_count / firing_count / pending_count /
state_summary / alerts (list with labels, annotations,
state, active_at, value). Enhanced with correlation
information when requested.
Inputs: instance, correlation_context, include_correlation_info. Returns: alerts, total_count, firing_count, pending_count, +1 more
prometheus_list_label_valuesARead-onlyIdempotent
List all values for a specific label from Prometheus.
Wraps GET /api/v1/label/{label_name}/values. Returns all distinct
values for the named label across all time series, optionally filtered
by a series selector.
Use this to discover what entities exist for a given label dimension — for example, which jobs are running, which instances are scraped, or which namespaces have metrics. This is essential for building targeted PromQL queries during investigation.
Examples:
- Use when: "What jobs does Prometheus scrape?"
→ label='job'; read the values list.
- Use when: "What instances are in the 'node-exporter' job?"
→ label='instance', match='{job="node-exporter"}'.
- Use when: "What namespaces have metrics?"
→ label='namespace'.
- Don't use when: You want metric names
(call prometheus_list_metrics — has substring filtering).
- Don't use when: You want current metric values
(call prometheus_query with a PromQL expression).
Returns:
dict with label / match / total_count /
returned_count / truncated / values (sorted list).
Inputs: label (required), match, instance. Returns: label, match, values, truncated, +2 more
prometheus_list_metricsARead-onlyIdempotent
List all metric names known to Prometheus, with optional substring filter.
Wraps GET /api/v1/label/__name__/values. Prometheus returns all metric
names at once — no pagination. Output is capped at 500 metrics after filtering,
with a truncation hint when more exist.
Use this first to discover valid metric names before writing PromQL
expressions for prometheus_query or prometheus_query_range.
Examples:
- Use when: "What metrics does Prometheus have about HTTP requests?"
→ pattern='http'; read the metrics list.
- Use when: "List all node_exporter metrics"
→ pattern='node_'.
- Use when: Starting a monitoring investigation — list metrics first
to discover what's instrumented, then query specific ones.
- Don't use when: You already know the exact metric name and want to
query its value (call prometheus_query directly — one fewer round trip).
- Don't use when: You want to see current alert state
(call prometheus_list_alerts).
Returns:
dict with total_count / returned_count / truncated /
pattern / metrics (sorted list).
Inputs: pattern, instance. Returns: metrics, pattern, truncated, total_count, +1 more
prometheus_list_rulesARead-onlyIdempotent
List recording and alerting rules from Prometheus.
Wraps GET /api/v1/rules. Returns rule groups with their rules,
including the PromQL expression, rule type (recording or alerting),
and for alerting rules their current state (firing/pending/inactive).
Use this to understand the alerting configuration, find recording rules that pre-compute useful aggregations, and investigate which rules are currently firing or have health issues.
Examples:
- Use when: "What alerting rules are configured?"
→ type='alert'; inspect rule names and expressions.
- Use when: "Are there any recording rules I can use instead of
computing aggregations from scratch?"
→ type='record'; look for rules matching your investigation.
- Use when: "Why is this alert firing? What's its PromQL expression?"
→ call with no filter; find the alert by name; read its query.
- Don't use when: You want to see which alerts are currently
firing (call prometheus_list_alerts — shows active alerts with
state and value, without the PromQL definition).
Returns:
dict with type_filter / total_groups / total_rules /
recording_count / alerting_count / groups (list of
rule groups with name, file, rule_count, rules).
Inputs: type, instance. Returns: groups, total_rules, type_filter, total_groups, +2 more
prometheus_list_targetsARead-onlyIdempotent
List Prometheus scrape targets, summarised by job and health.
Wraps GET /api/v1/targets. Returns scrape targets with job name,
instance address, health status (up / down / unknown),
last scrape duration in milliseconds, and any last error. Also returns
a summary grouped by job and health state.
Examples:
- Use when: "Which targets are currently down?"
→ filter targets where health='down' and check last_error.
- Use when: "How many instances of the 'node-exporter' job are up?"
→ check job_summary for the 'node-exporter' entry.
- Use when: Investigating a scrape failure — list targets for the
affected job to see which instances have errors.
- Don't use when: You want metric values from a target
(call prometheus_query with label matchers instead).
- Don't use when: You want alert status
(call prometheus_list_alerts instead).
Returns:
dict with state_filter / total_count / up_count /
down_count / unknown_count / job_summary (per-job health counts) /
targets (list with job, instance, health,
last_scrape_duration_ms, last_error, labels).
Inputs: state, instance. Returns: targets, up_count, down_count, job_summary, +3 more
prometheus_map_service_dependenciesARead-onlyIdempotent
Map service-to-service dependencies from live Prometheus traffic metrics.
Aggregates rate(<metric>[<window>]) by (source_label, destination_label)
via query_range and turns each returned series into a directed
dependency edge (nodes = services, edge weight = mean request rate). No
simulation — every node/edge comes from real series. When error_matcher
is supplied a second query derives a per-edge error rate.
Use this to:
Discover which services call which, and how heavily.
Spot high-error dependencies during an incident.
Examples:
- "Map dependencies for the prod namespace over the last 15 minutes"
→ matchers={"namespace": "prod"}, time_window="15m".
- "Which edges are erroring?"
→ error_matcher='response_code=~"5.."'.
Returns:
ServiceDependencyMapResult with nodes and directed edges
(sorted by throughput), plus the query parameters used.
Inputs: step, metric, instance, matchers, +4 more. Returns: step, edges, nodes, metric, +7 more
prometheus_queryARead-onlyIdempotent
Execute an instant PromQL query against Prometheus.
Wraps GET /api/v1/query. Returns the result type (vector, scalar,
matrix, string) and a list of samples each carrying labels, timestamp,
and value. For vector results each element is one time series at the
evaluation instant.
Examples:
- Use when: "Is the payment service up right now?"
→ query='up{job="payment-service"}'.
- Use when: "What is the current HTTP request rate?"
→ query='sum(rate(http_requests_total[5m])) by (job)'.
- Use when: "Show me all metrics for a specific instance"
→ query='{instance="localhost:9090"}'.
- Don't use when: You want to see how a metric changed over time
(call prometheus_query_range with start/end/step).
- Don't use when: You don't know the metric name yet
(call prometheus_list_metrics first to discover names).
Returns:
dict with query / time / result_type / result_count /
data (list of samples with labels, timestamp, value).
Inputs: query (required), time, instance. Returns: data, time, query, result_type, +1 more
prometheus_query_rangeARead-onlyIdempotent
Execute a PromQL range query returning time-series data points.
Wraps GET /api/v1/query_range. Returns one series per matching time
series, each with labels and a list of [timestamp, value] pairs.
Total points across all series are capped at 5000 with a truncation hint.
Prometheus may reject the query with HTTP 422 (bad_data) if the step produces too many data points (> 11,000 per series). Increase the step or narrow the time range if this happens.
Note: The Prometheus API does not support filtering by branch or commit in this endpoint — filters are expressed purely in PromQL label matchers.
Examples:
- Use when: "Show me CPU usage over the last hour with 1-minute resolution"
→ query='rate(node_cpu_seconds_total[5m])', step='1m'.
- Use when: "Graph HTTP error rate for the last 24 hours"
→ query='rate(http_requests_total{status=~"5.."}[5m])',
start='2024-01-15T00:00:00Z', end='2024-01-16T00:00:00Z',
step='5m'.
- Use when: Investigating a past incident — pick the time window of the
incident and use a fine step.
- Don't use when: You only want the current value
(call prometheus_query — faster and simpler).
- Don't use when: You want alert history (call prometheus_list_alerts).
Returns:
dict with query / start / end / step / result_type /
series_count / total_points / truncated /
data (list of series with labels, point_count, values).
Inputs: end (required), step (required), query (required), start (required), +1 more. Returns: end, data, step, query, +5 more
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 21 tools
Most tools map cleanly to a distinct API endpoint, and descriptions explicitly cross-reference siblings (prometheus_list_alerts vs alertmanager_list_alerts, query vs query_range). The main ambiguity is among the three alert analytics tools (correlate_alerts_across_instances, group_alerts_by_service, detect_cascading_alerts), which all describe related-cluster/cascade analysis over the same alert data and rely on prose rather than distinct inputs to steer selection.
The dominant pattern is subsystem-prefixed snake_case verb_noun (prometheus_query, prometheus_list_targets, alertmanager_list_silences, federation_list_instances), which is very predictable. Deviations are the three unprefixed verb-first analytics tools (correlate_alerts_across_instances, group_alerts_by_service, detect_cascading_alerts), which break the prefix convention but remain readable.
21 tools is on the heavy side for a read-only monitoring proxy, though it spans three real domains (Prometheus API, Alertmanager API, federation/health). Several introspection tools (build_info, runtime_info, cardinality, health_check, metadata) are each thin wrappers around a single endpoint and could plausibly be consolidated.
Coverage is strong for read/diagnostic workflows: instant and range queries, metadata, label, metric and target discovery, rules, alerts, TSDB cardinality, runtime/build info, health, plus Alertmanager status, alerts, silences, and groups. Gaps are mostly write/lifecycle operations (creating or expiring silences, modifying rules) that an investigation-focused agent could want during incident mitigation.
Maintenance
Related MCP Connectors
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Uptime, SSL, DNS and domain monitoring you can talk to from Claude or any MCP client.
MCP observability. Query live traffic, errors, duration, and alerts from your AI agent.
The official Model Context Protocol server for Ambee. It gives any MCP-compatible AI assistant — Claude, ChatGPT, Cursor, VS Code, Ollama, and more direct access to live air quality, pollen, and weather data. To get started, including information on signing up and obtaining your Ambee key, check out the Ambee documentation on https://docs.ambeedata.com
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to query Prometheus metrics, monitor alerts, and analyze system health through read-only access to your Prometheus server with built-in query safety and optional AI-powered metric analysis.MIT
- AlicenseNot gradedqualityDmaintenanceA lean MCP server that provides LLM agents with transparent access to multiple Prometheus instances for metrics analysis and SRE operations.4GPL 2.0
- AlicenseBqualityDmaintenanceEnables interacting with Prometheus through MCP for querying metrics, series, alerts, rules, and server status using natural language.9224 npm6MIT

prometheus-mcp-serverofficial
AlicenseNot gradedqualityCmaintenanceMCP server that lets LLMs interact with a running Prometheus instance via its API, enabling metric analysis, PromQL queries, and rule/target inspection. It also provides embedded documentation and guided runbooks for common Prometheus tasks.142Apache 2.0