@cyanheads/devops-status-mcp-server
Check vendor status pages, inspect SSL/TLS certificates, verify DNS propagation, and get incident-response playbooks via MCP. STDIO or Streamable HTTP.
7 Tools • 1 Resource
Overview
Status pages for 71 vendors — AWS, Google Cloud, Azure, Slack, and more — in one normalized format. Check who's down, pull incident timelines, watch your stack, and get an incident-response playbook. Plus TLS certificate and DNS checks for any domain. No API keys required.
Tools
Resources
All resource data is also reachable via tools. Tool-only agents are fully supported.
Capability reference
devops_list_vendors tool
- Optional free-text
query (name and slug, case-insensitive) and optional category: cloud, cdn-edge, dev-platform, data, comms, auth, monitoring, ai
- Returns slug (what to pass to other tools), display name, category, and status page URL; an empty result carries a
notice naming what missed
- 71 built-in entries, most on Atlassian Statuspage;
fly-io, planetscale, supabase, brevo, clerk, openai, elevenlabs, and cohere (incident.io-hosted pages, read through their own page feed); aws, gcp, azure, azure-devops, gitlab, neon, slack, redis-cloud, and the nine Datadog site entries (datadog for US1, datadog-us3, datadog-us5, datadog-eu1, datadog-uk1, datadog-ap1, datadog-ap2, datadog-us1-fed, datadog-us2-fed) use native-API adapters. Statuspage-compatible pages not in the registry work by passing a raw base URL to other tools
- These Atlassian products have their own entry (
jira, confluence, bitbucket, jira-service-management, atlassian-compass, atlassian-developer, trello, opsgenie, atlassian-statuspage, atlassian-guard). atlassian reads the umbrella page, which publishes no components and only cross-product incidents; any tool serving it says so in notice and names those slugs
Built-in vendor registry:
devops_status_check tool
- Accepts vendor slugs (e.g.,
github, aws) or raw Atlassian Statuspage base URLs, mixed freely; a URL is read at its origin and path. A URL on the status page of a vendor read through another backend (https://status.gitlab.com) gets an error naming its slug, and a URL on an incident.io page reads that page's Statuspage-compatible API, which lists 25 components and 25 incidents at most, with no component groups and maintenance windows without their times — pass the slug (openai) for the full feed. Up to 20 are checked per call; the rest come back in deferred_vendors with a notice to resend them. More than 100 entries is rejected (invalid_arguments)
mode: "summary" (default) returns the indicator, degraded components, and active incidents; mode: "detailed" adds the full component list (component_limit, default 50, clamped to 500) and scheduled maintenance windows. component_filter matches component or group names
- A list longer than
component_limit pages: each capped vendor returns next_component_offset, which you pass back as component_offset (applied after component_filter). degraded_components and active_incidents are complete on every page
- One failing vendor never blocks the rest (per-vendor
error field). Results are cached for 60 seconds (cached: true), and summary counts operational / degraded / down / maintenance / unavailable
indicator is the vendor's own rating, and a Statuspage page can keep it at none while an incident it rated minor or worse is open. summary counts each vendor at the worse of the two; incidents rated none or maintenance and scheduled windows never count as an outage
nextToolSuggestions pre-fills a devops_suggest_action call for each vendor with an active problem — the vendors counted degraded or down
devops_get_incidents tool
filter: all (default), active, resolved, or scheduled (maintenance windows only). limit (default 20, max 50) with offset for paging
- Returns per-update bodies as plain text in chronological order, affected component names, duration in minutes for resolved incidents, and a shortlink. Google Cloud and AWS incidents also carry
affected_locations (region ID, name, and whether it is still affected; on AWS, the one region the event names, or global). Each incident carries source: api, or history for an archive record (title, impact, times, and final message only)
since and until (YYYY-MM-DD, UTC) bound the start date and work with filter: "all" or "resolved"; since also reads the Statuspage history archive back up to 24 months, past the API's newest-50 limit; a call that hits that limit without since is told to set it. On incident.io vendors the feed lists the last 90 days, a notice says when the page publishes older incidents, and since reads them back up to 24 months as full records. match keeps incidents whose title, component names, or (on Google Cloud and AWS) location names and IDs contain every space-separated term given (up to 200 characters); a term joined by punctuation, such as eu-central-1, must find its words together and in order within one of them
- Responses are capped at 64,000 bytes per surface; a cut page sets
truncated: true and nextOffset. full_updates: true keeps every update on a single oversized incident
- Errors:
invalid_since and invalid_until (filter: "active" or "scheduled", or until before since); statuspage_not_found (a raw URL that serves no Atlassian Statuspage API, naming the registry slug where one exists; not retryable, unlike statuspage_unavailable)
- Feeds with no data for a filter return empty:
resolved for Azure, scheduled for AWS, Azure, Azure DevOps, Google Cloud, and Slack, and for a Statuspage page that publishes no maintenance feed (a notice says so). An empty active or resolved that read no incident suggests scheduled only where the vendor publishes maintenance windows. History is short where a vendor reads no archive: AWS keeps resolved events for a few hours, and Status.io (gitlab, neon) lists current incidents only. On those feeds a since earlier than the oldest record returned, or an until without since, gets a notice naming the date the feed starts
devops_watch_stack tool
- Pass
vendors (up to 20) once to save the stack under stack_name (1-64 characters; several stacks can coexist, e.g. "production"); later calls can omit vendors. Over 20 is rejected with stack_too_large. Saving works on stdio and on HTTP with MCP_AUTH_MODE=jwt|oauth; with the default in-memory storage, saved stacks do not survive a restart
- On HTTP with
MCP_AUTH_MODE=none nothing is saved: a call with vendors runs the sweep and returns stack_persisted: false, and a call without vendors fails with vendors_required
- Aggregate
health: all_operational / maintenance / degraded / partial_outage / major_outage / unknown (a vendor could not be reached), rolled up from the same per-vendor rating as summary. component_limit and nextToolSuggestions work as in devops_status_check; there is no component_offset, so a capped vendor's next_component_offset continues in devops_status_check
devops_check_certs tool
- Accepts hostnames, IP addresses,
host:port, or URLs; a port on the entry overrides port (default 443). The first 10 entries are inspected (up to 100 accepted); the rest come back in deferred_domains to resend
- Reports days to expiry (
warning under 30 days, critical under 7), subject and SANs, issuer, chain depth, TLS version (1.0 and 1.1 flagged insecure), cipher suite, and HSTS on HTTP ports
status: "critical" separates a hostname mismatch (hostname_verification_error) from an untrusted chain (authorization_error). Per-domain failures come back inline as status: "error" with the reason in error
timeout_ms is one deadline per domain (max 15000)
devops_check_dns tool
- Accepts domain names, URLs, or
host:port; the first 10 domains are queried (up to 100 accepted) and the rest come back in deferred_domains. IP addresses get their own error row
- Queries Google (8.8.8.8), Cloudflare (1.1.1.1), and Quad9 (9.9.9.9) in parallel for A, AAAA, CNAME, MX, TXT, NS, CAA, and SOA (default A, AAAA, MX, TXT)
- Reports per-resolver latency and typed discrepancies:
partial_resolution (some resolvers did not answer) is a real problem; value_variation (different values) is normal for anycast or geo-steered domains
- Custom
resolvers take public IP literals, first 5 queried and the rest in deferred_resolvers; private and loopback addresses need DEVOPS_STATUS_ALLOW_PRIVATE_TARGETS=true. timeout_ms applies per domain and resolver (max 10000)
devops_suggest_action tool
- Takes a vendor slug or display name and returns a category-tailored markdown playbook, with generic guidance for unrecognized vendors
- Optional
incident_summary, affected_components, and vendor_indicator add a subsystem section and severity-based urgency
nextToolSuggestions pre-fills follow-up calls, including cert and DNS checks when your_domain is given. With DEVOPS_STATUS_DISABLE_ACTIVE_PROBES=true, guidance uses manual commands (dig, openssl s_client) instead
devops-status://vendors/{name} resource
- Returns the full registry entry for a vendor slug: status page URL, category, API type (
statuspage, statusio, slack, aws, gcp, azure, azure-devops, firehydrant, datadog, incidentio), component_scope_prefix where a vendor covers part of a shared page (sendgrid), and product_slugs on an umbrella page that publishes no components (atlassian)
- Cached publicly for 1 hour; the same data is reachable via
devops_list_vendors
Features
Built on @cyanheads/mcp-ts-core: stdio and Streamable HTTP transports, pluggable auth (none / jwt / oauth), swappable storage (in-memory, filesystem, Supabase, Cloudflare KV/R2/D1), structured logging with optional OpenTelemetry tracing.
DevOps-status-specific:
- No API keys required — every status backend is a public API; TLS and DNS use the Node.js standard library
- 71-vendor built-in registry across eight categories; native adapters normalize incident.io, Status.io, Slack, AWS Health, Google Cloud, Azure, Azure DevOps, Firehydrant, and Datadog into the Statuspage shape
- Vendor-authored text is escaped so it reads as data, not markup; DNS record values and certificate names render as code spans that read back exactly as resolved
- 60-second in-memory cache on status reads, shared across tenants; concurrent reads of one URL share a single request
devops_watch_stack saves named vendor lists for repeat checks and pre-deploy sweeps
devops_suggest_action is deterministic, with no LLM sampling dependency
Agent skill:
devops-status-guide carries the gates for using these tools together: what a status reading can and cannot establish, how far incident history reaches on each backend, where regions live, and how to diagnose a DNS change or a certificate failure. MCP clients read it from the server as skill://devops-status-guide/SKILL.md; it also ships in the npm package, the Docker image, and the .mcpb bundle, and plugin installs scan it from skills/.
Agent-friendly output:
- Batch tools never let one failing target block the rest; errors come back as inline
error fields
checked_at on every status result is the time of the check; cached: true marks a status read served from the cache, which can be up to the cache TTL (60 seconds by default) older
- Fixed indicator and status enums (
none / minor / major / critical / maintenance) let callers branch on data
nextToolSuggestions pre-fills follow-up tool arguments from incident context
Getting started
Connect to the hosted instance, no install:
Or run it locally:
For other clients, add this to the MCP config file:
No API key required. For Docker or a source build, see Running the server.
Configuration
No API keys required. All environment variables are optional.
An invalid DEVOPS_STATUS_* value stops startup with an error naming the variable.
See .env.example for the full list of optional overrides.
Running the server
Local development
Requires Bun v1.4.4 or higher (or Node.js v24+).
-
Clone and install:
-
Build and run:
-
Run checks and tests:
Docker
The image defaults to HTTP transport, stateless session mode, and logs to /var/log/devops-status-mcp-server. To run it over stdio from an MCP client, use "command": "docker" with "args": ["run", "-i", "--rm", "-e", "MCP_TRANSPORT_TYPE=stdio", "ghcr.io/cyanheads/devops-status-mcp-server:latest"]. To build locally, docker build -t devops-status-mcp-server .; OpenTelemetry peer dependencies are installed by default — build with --build-arg OTEL_ENABLED=false to omit them.
Project structure
Development guide
See CLAUDE.md for development guidelines and architectural rules. The short version:
- Handlers throw, framework catches — no
try/catch in tool logic
- Use
ctx.log for request-scoped logging, ctx.state for tenant-scoped storage
- Register new tools and resources via the barrels in
src/mcp-server/*/definitions/index.ts
devops_check_certs and devops_check_dns use only Node.js stdlib — add no external deps for these paths
Contributing
Issues are welcome. Run checks and tests before submitting:
License
Apache-2.0 — see LICENSE for details.