Agents

Set up the down.guide MCP server for me. down.guide is a free, public outage feed for 400 cloud and SaaS providers: no login, no API key.

- Server name: down-guide
- Transport: Streamable HTTP
- URL: https://mcp.down.guide/mcp

Add it the way you normally add MCP servers (your CLI command or your MCP config file), then confirm it connects and lists its two tools, search and execute. To finish, use them to run the stack.check operation on this project's package.json or lockfile and tell me which of my providers have a problem right now.

incident.list

The newest incidents across providers

incident.get

One incident by id: a down.guide id (DWN-…)

provider.list

The provider catalog with live state; q matches names and category narrows it

provider.get

One provider by id: the provider's slug or name, such as github or aws

cascade.list

Upstream outages and the providers that broke with them, newest first; provider names the upstream

cascade.get

One cascade by id: the cascade's slug, such as aws-2026-11-06

entities.search

Search incidents

feed.read

The items a /feed subscriber gets, as JSON Feed, newest first

stack.check

Which providers a project depends on, and whether any has a problem now

dependency.check

One dependency to the providers behind it and their live state

leaderboard.get

The Down Index

feedback.send

Send a note to the people who run down.guide

report.outage

Report that a provider isn't working for you (or for the agent's user) right now

report.affected

Say an existing incident affects you

Get provider

GET /api/v1/providers/<slug>

List feed

GET /api/v1/incidents

Get incident

GET /api/v1/incidents/<id>

Get cascade

GET /api/v1/cascades/<slug>

Search

GET /api/v1/search

Check stack

POST /api/v1/check

Check dependency

GET /api/v1/dependency

Get leaderboard

GET /api/v1/leaderboard

Send feedback

POST /api/v1/feedback

Report outage

POST /api/v1/reports

Report affected

POST /api/v1/reports/affected

Catalog

GET /api/v1/providers

Every provider, one line each

GET /api/v1/status

curl https://down.guide/api/v1/providers/github

llms.txt · llms-full.txt · openapi.json · agent.json

The MCP server lists exactly two tools that never change: search finds an operation and execute runs one. Every capability is an operation (provider.get, incident.list, stack.check, entities.search, <type>.list and <type>.get, with a REST twin each), so a new one never changes the tool list. The old tools (get_provider, list_feed and the rest) stay callable and are no longer listed: deprecated since 2026-11-11, removed from these docs on 2027-01-10.

Every entity type has a REST list and get in one shape (https://down.guide/openapi.json lists them): /api/v1/incidents, /providers and /cascades answer { items, next_cursor, total? } and every object carries type (incident, provider or cascade), kind, id, url and title. /api/v1/cascades is new, with limit (1 to 200, default 30) and cursor, and a parameter that is unknown or does not apply is a 400. /api/v1/providers takes limit, cursor and q, and without a limit still returns the whole catalog in one answer, providers beside items.

Every incident carries confidence (confirmed or corroborated), confidenceLabel, acknowledged (true when the provider's own page posted it, from acknowledgedAt), providerStatus and cascade, in incident JSON, feeds, search and MCP results. The labels are the same words as cve.now: Confirmed and Corroborated.

confidence:, acknowledged: and cascade: are search keys, and confidence is a typed parameter of search, list_feed and /api/v1/incidents. confidence:unverified is valid and matches nothing: unverified evidence is never public.

The get_cascade MCP tool and GET /api/v1/cascades/<slug> return one cascade (an upstream outage and the providers that broke with it, with the incident behind each); an incident's cascade field is the slug. The CLI has the same as down.guide cascade <slug>.

Search is one search, the same on cve.now and down.guide. type: (and the type parameter) takes entity types: incident, provider and cascade, or all; date: is the shared name for started:, and filter:major is the priority slice /feed?filter=major reads. The response's items hold the matching incidents and the providers and cascades the query names, each tagged with its type and an exact name match first; results holds the incidents alone, as before. facets gains type.

Regions come at three levels: region: a cloud provider's own region, as the provider names it (us-east-1, europe-west4); area: a market area (us, eu, apac, latam, me); country: one country (au, ca, in, jp).

level (each incident)

Now confidence (confirmed or corroborated) and acknowledged; not_acknowledged is corroborated with acknowledged false, and signal is unverified and never served

levelLabel (each incident)

Now confidenceLabel (Confirmed or Corroborated)

status (each incident)

Now providerStatus (an update's own status keeps its name)

event (each incident)

Now cascade

facets.level (search)

Now facets.confidence

level: (search)

Now confidence: (confirmed, corroborated or unverified) and acknowledged: (true or false); level:not_acknowledged is acknowledged:false

event: and has:event (search)

Now cascade: and has:cascade

level (search, list_feed and /api/v1/incidents (/feed?level= keeps working))

Now the confidence parameter