incidentrelay-mcp
MCP server for IncidentRelay, exposing its entire REST API (279 operations) through gateway and curated tools for on-call scheduling, alert routing, and incident management.
README
incidentrelay-mcp
An MCP server for IncidentRelay — self-hosted on-call scheduling, alert routing, and incident management.
It exposes IncidentRelay's entire REST API (279 operations) to MCP clients (Claude Desktop, Claude Code, Cursor, …) through two layers:
- 3 gateway tools that make every operation reachable, driven by the server's own OpenAPI document.
- 25 curated tools that wrap the common on-call workflows (acknowledge/ resolve alerts, incidents, on-call schedule, silences, maintenance windows) with typed, ergonomic inputs.
The server never modifies IncidentRelay. It is a pure REST client that authenticates with a bearer token you provide.
Run
No install needed — run it with npx (Node.js ≥ 20):
# from GitHub (works today)
IR_BASE_URL=https://incidentrelay.example.com IR_TOKEN=<token> \
npx -y github:NikGariel/IncidentRelay-mcp
# from npm (after the package is published)
IR_BASE_URL=https://incidentrelay.example.com IR_TOKEN=<token> \
npx -y incidentrelay-mcp
npx builds the package automatically on first run (a prepare script compiles
TypeScript), so no separate build step is required.
From source
npm install # runs the build via the prepare script
npm start
Configure
The server needs the IncidentRelay base URL and a bearer token (a JWT access token or a personal API token created in IncidentRelay under Profile → API tokens, with the scopes you want to allow).
| Setting | Env | Flag | Required | Default |
|---|---|---|---|---|
| Base URL | IR_BASE_URL |
--base-url |
yes | — |
| Bearer token | IR_TOKEN |
--token |
yes¹ | — |
| Transport | IR_TRANSPORT |
--transport |
no | stdio |
| HTTP port | IR_PORT |
--port |
no | 3000 |
| HTTP host | IR_HOST |
--host |
no | 127.0.0.1 |
| Read-only | IR_READONLY |
--readonly |
no | false |
| Request timeout (ms) | IR_TIMEOUT_MS |
— | no | 30000 |
| OpenAPI URL override | IR_OPENAPI_URL |
— | no | {base}/api/openapi.json |
| TLS verification | IR_VERIFY_TLS |
— | no | true |
¹ The token is optional only if you never call write operations.
At startup the server fetches {IR_BASE_URL}/api/openapi.json (a public
endpoint) to build its operation catalog, so it always matches the deployed
IncidentRelay version. If that fetch fails, it falls back to a bundled
snapshot and logs a warning.
Use with Claude Desktop / Claude Code (stdio)
Add to your MCP client config (e.g. claude_desktop_config.json).
Recommended — run the prebuilt binary directly (starts in <0.5s, so it never hits the client's startup timeout):
{
"mcpServers": {
"incidentrelay": {
"command": "node",
"args": ["/absolute/path/to/incidentrelay-mcp/dist/index.js"],
"env": {
"IR_BASE_URL": "https://incidentrelay.example.com",
"IR_TOKEN": "<personal-api-token>"
}
}
}
}
Build it once with npm install && npm run build (or npm install -g .).
Alternative — npx. Once published to npm:
{ "command": "npx", "args": ["-y", "incidentrelay-mcp"], "env": { "…": "…" } }
⚠️ Avoid
npx -y github:…in a client config: its first run clones the repo, installs dependencies, and compiles TypeScript, which can exceed an MCP client's ~30s connection timeout on a cold cache. Use the prebuilt binary (or the npm form after publishing) for a client that spawns the server on demand.
The server itself starts instantly regardless of network: it loads a bundled OpenAPI snapshot synchronously, then refreshes the catalog from the live server in the background.
Use over HTTP (Streamable HTTP)
IR_BASE_URL=https://incidentrelay.example.com IR_TOKEN=... \
npx -y github:NikGariel/IncidentRelay-mcp --transport http --port 3000
# serves MCP at http://127.0.0.1:3000/mcp
The HTTP transport is stateless (one server per request), binds to
127.0.0.1 by default, and has DNS-rebinding protection enabled.
Tools
Gateway (cover all 279 operations)
| Tool | Purpose |
|---|---|
ir_list_operations |
Search/list the API catalog by query, tag, or method. |
ir_describe_operation |
Show one operation's params and request-body schema. |
ir_call |
Execute any operation by operationId (or method+path). |
Typical flow: ir_list_operations → ir_describe_operation → ir_call.
Curated (common on-call workflows)
Alerts: ir_list_alerts, ir_get_alert, ir_ack_alert, ir_resolve_alert,
ir_comment_alert ·
Incidents: ir_list_incidents, ir_get_incident, ir_create_incident,
ir_add_incident_responder, ir_set_incident_priority ·
On-call: ir_who_is_on_call, ir_oncall_health, ir_list_rotations,
ir_create_rotation_override ·
Silences: ir_list_silences, ir_create_silence, ir_remove_silence ·
Maintenance: ir_list_maintenance_windows, ir_create_maintenance_window ·
Context: ir_list_teams, ir_list_routes, ir_list_channels, ir_whoami ·
Heartbeats: ir_list_heartbeats, ir_send_heartbeat.
Safety
- Confirm-guard. Destructive operations (
DELETE, and paths withdisable/remove/delete/reset/regenerate-token/intake-token/cancel/merge/revoke) do not run unless you passconfirm: true. Without it, the tool returns a dry-run preview. - Read-only mode. Set
IR_READONLY=trueto refuse every non-GEToperation at the tool layer. - Token redaction. Bearer tokens are stripped from logs and error output.
Security
IR_VERIFY_TLS=false disables TLS certificate verification for outbound calls
to IncidentRelay. This exposes traffic to man-in-the-middle attacks and should
only be used on a trusted network. For self-hosted self-signed certificates,
prefer adding the server's CA to your trust store (or use a properly-issued
certificate) instead of disabling verification. When disabled, the server
prints a warning at startup.
Development
npm test # run the vitest suite
npm run dev # run from source with tsx
npm run build # typecheck + emit dist/ + copy the OpenAPI snapshot
Regenerating the bundled OpenAPI snapshot: see
scripts/generate-snapshot.md.
Design & plan
- Design spec:
docs/superpowers/specs/2026-08-17-incidentrelay-mcp-design.md - Implementation plan:
docs/superpowers/plans/2026-08-17-incidentrelay-mcp.md
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。