Skip to main content
Glama

mcp-allquiet

CI PyPI License

An MCP server for All Quiet, the incident management and on-call platform. It lets Claude, Cursor or any MCP client triage incidents, check who is on call and reach every operation of the All Quiet public API.

Built on the official MCP Python SDK v2 (mcp 2.x). Not affiliated with All Quiet.

Why

All Quiet has a complete public API (136 operations) but no official MCP server. Exposing one tool per operation would put 136 tool schemas into the model's context before the first question. This server ships 10 tools:

  • 5 incident tools for the on-call loop, returning compact summaries.

  • 5 generic tools driven by the bundled OpenAPI spec, which reach every other operation (teams, schedules, overrides, integrations, routing, status pages, maintenance windows...).

Related MCP server: pagerduty-mcp

Install

Requires uv and an All Quiet API key (the API needs a Pro or Enterprise plan).

Claude Code:

claude mcp add allquiet -e ALLQUIET_API_KEY=your-key -- uvx mcp-allquiet

Claude Desktop, Cursor and other clients (mcpServers JSON):

{
  "mcpServers": {
    "allquiet": {
      "command": "uvx",
      "args": ["mcp-allquiet"],
      "env": { "ALLQUIET_API_KEY": "your-key" }
    }
  }
}

Configuration

Variable

Required

Default

ALLQUIET_API_KEY

yes

Organization API key or personal access token

ALLQUIET_BASE_URL

no

https://allquiet.app

Use https://allquiet.eu for the EU region. Must be https://

The API key decides what the model can do

All Quiet keys carry per-resource scopes (incidents:list, teams:update...), can be limited to teams, and can expire. The server has no read-only switch on purpose: create a key with only list/get scopes and every write fails at the API. A personal access token can never do more than its user can in the UI.

A personal access token only reaches teams its user belongs to. A team created through the API stays out of reach (403) until a POST /team-membership adds the user as Administrator. Organization API keys skip team role checks.

Keys spanning several organizations need the organizationId query parameter on many operations. The API says so in its error, and the model can pass it through the generic tools.

Tools

Tool

What it does

Kind

list_incidents

Filter by status, severity, team, user, text, unattended, date. Compact summaries, paged

read

get_incident

Full incident as markdown: timeline, attributes, assignees

read

create_incident

Open an incident

write

update_incident

Record an action (resolve, acknowledge, comment...) and/or change severity

write

who_is_on_call

Current (or past) on-call per team, with escalation tier

read

list_operations

Search the 136 API operations, with the key scope each needs

read

describe_operation

Parameters and request body schema of one operation

read

call_read

GET any operation

read

call_write

POST / PUT / PATCH any operation (JSON bodies)

write

call_delete

DELETE any operation

destructive

Tools carry MCP annotations (readOnlyHint, destructiveHint), so clients can ask before writes and deletes.

Security

  • The key is read from the environment and only sent to the configured https:// host. It never appears in tool output or error messages (tested).

  • Generic tools accept only path templates that exist in the bundled spec. Path parameters are percent-encoded as a single segment, so a value cannot rewrite the path.

  • Responses are truncated at 40k characters and error bodies at 2k.

  • Incident content is untrusted input. Titles, messages and attributes often come from alert payloads written by third parties. Treat what the model reads there as data, and keep write scopes off keys you use with untrusted incident sources.

Report vulnerabilities through SECURITY.md.

Limitations

  • stdio transport only.

  • POST /inbound-integration/{id}/call-routing-media/{purpose} needs multipart/form-data and is not supported.

  • No retries. A 429 returns the Retry-After value to the model.

Development

uv sync
uv run pytest
uv run ruff check && uv run ruff format --check && uv run ty check

Try it with the MCP Inspector:

ALLQUIET_API_KEY=your-key npx @modelcontextprotocol/inspector uv run mcp-allquiet

Refresh the bundled OpenAPI spec:

curl -s https://allquiet.app/api/swagger/public-v1/swagger.json \
  | python -c "import json,sys; json.dump(json.load(sys.stdin), sys.stdout, separators=(',',':'))" \
  > src/mcp_allquiet/openapi.json

See CONTRIBUTING.md.

License

MIT

Available Tools

10 tools
call_deleteA
Destructive

DELETE through any All Quiet API operation. Irreversible.

Inputs: path (required), query, path_params.

call_readA
Read-only

GET any All Quiet API operation. Find it first with list_operations.

Inputs: path (required), query, path_params.

call_writeB

Create or change something through any All Quiet API operation (JSON bodies only).

Inputs: path (required), method (required), body, query, +1 more.

create_incidentC

Open an incident in one of the key's teams.

Inputs: title (required), severity (required), status, message, +1 more.

describe_operationA
Read-only

Show an operation's parameters and JSON request body schema, with all $refs inlined. A schema used more than once appears in full once, tagged "$name", then as {"$see": name}.

Inputs: path (required), method (required).

get_incidentA
Read-only

Full incident as markdown (timeline, attributes, assignees) plus the intents update_incident can record on it now.

Inputs: incident_id (required).

list_incidentsA
Read-only

List incidents as compact summaries. Use get_incident for the full timeline.

Inputs: limit, offset, statuses, team_ids, +5 more.

list_operationsA
Read-only

List All Quiet API operations as METHOD path — summary [required key scope].

Inputs: query.

update_incidentB

Record an action (resolve, acknowledge, comment...) and/or change severity.

Inputs: incident_id (required), intent, message, severity.

who_is_on_callC
Read-only

Who is on call, per team, with escalation tier.

Inputs: team_ids, user_ids, timestamp.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 10 tools

    Disambiguation4/5

    The generic call_read/call_write/call_delete tools overlap somewhat with the specific incident tools, but their descriptions clearly distinguish them as raw API access versus curated workflows. The incident-specific tools are distinct and well-scoped.

    Naming Consistency4/5

    Most tools follow a consistent verb_noun pattern (list_operations, describe_operation, call_read, list_incidents, get_incident, create_incident, update_incident). The generic call_* tools deviate slightly but still follow a predictable verb style.

    Tool Count4/5

    10 tools is a reasonable count for an API wrapper server. The generic call_* tools plus specific incident tools create slight redundancy but the count is well within acceptable range.

    Completeness4/5

    The server covers incident lifecycle (list, get, create, update) plus on-call lookup and raw API access for anything else. Minor gaps exist (no explicit delete_incident tool) but the generic call_delete covers that.

    Maintenance

    ActivityStale
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables managing incident response workflows including incidents, on-call, postmortems, and more through the Runframe platform using MCP tools.
      45 npm
      4
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables PagerDuty incident response operations including listing incidents, acknowledging and resolving incidents, looking up on-call schedules, and listing services.
      MIT
    • A
      license
      B
      quality
      F
      maintenance
      Enables management of PagerDuty incidents, services, schedules, and more directly from MCP-enabled clients, with embedded interactive UIs for incident command center, on-call management, and other features.
      63
      79
      Apache 2.0