接入你的智能体
從發布者安裝。不是 PlugMCP 端點。
遠端 URL 還是本機指令? 兩種安裝方式的差異
Docs · Coroot · Changelog · Contributing · MCP manifest
Coroot MCP is a Model Context Protocol (MCP) server that turns your Coroot observability stack into a set of well-typed tools an LLM assistant can call for root-cause analysis.
This is a community-maintained integration and is not an official Coroot product or endorsement.
It is implemented as a Spring Boot 3 / Spring AI 1.1.x application and exposes a JSON-RPC 2.0 MCP endpoint over HTTP at /mcp.
The project is licensed under the MIT License.
You can try the MCP server without a Coroot instance by enabling the built-in stub client profile. This returns synthetic incidents and health snapshots that are good enough for testing tool wiring.
./mvnw spring-boot:run -Dspring-boot.run.profiles=stub-coroot
The MCP JSON-RPC endpoint will be available at:
POST http://localhost:8080/mcpTo point coroot-mcp at a real Coroot instance:
Create a local .env.local file in this repo (gitignored) with:
COROOT_API_URL=https://coroot.your-company.com
COROOT_API_KEY=<your Coroot API key>
COROOT_DEFAULT_PROJECT_ID=production
OPENAI_API_KEY=<your OpenAI/OpenAI-compatible key>
OPENAI_MODEL=gpt-4.1-mini
Start the server:
set -a
source .env.local
set +a
./mvnw spring-boot:run
Verify it is running:
curl -s http://localhost:8080/actuator/health
curl -s http://localhost:8080/mcp \
-H 'Content-Type: application/json' \
-d '{"jsonrpc":"2.0","id":"init-1","method":"initialize","params":{}}'
Point your MCP-aware client (Codex, Claude, Toolhive, etc.) at http://localhost:8080/mcp and start using tools like list_recent_incidents and investigate_incident.
This project speaks MCP over HTTP. Any MCP-aware coding assistant can talk to it once you point the client at the /mcp endpoint.
Assuming coroot-mcp runs on http://localhost:8080/mcp, add an MCP server entry in your Codex configuration, for example:
[mcp_servers.coroot-mcp]
command = "npx"
args = ["-y", "mcp-remote", "http://localhost:8080/mcp", "--allow-http", "--transport", "http-first"]
Restart Codex CLI and list MCP servers to confirm that coroot-mcp is available.
If you use a Claude-based environment that supports HTTP MCP servers, configure a new MCP server named coroot-mcp with:
http://localhost:8080/mcpYou can then call list_recent_incidents and summarize_incident_root_cause from within that environment.
Some tools and IDEs prefer JSON-based MCP configuration files. The snippets below mirror what many MCP-aware agents expect as of late 2025; adjust paths and secrets to your environment.
.mcp.json (project-level)You can keep a project-scoped MCP configuration in .mcp.json at the root of your repo:
{
"mcpServers": {
"coroot-mcp": {
"type": "http",
"url": "http://localhost:8080/mcp"
}
}
}
Some clients also support an env block here; if yours does, you can document expected variables:
{
"mcpServers": {
"coroot-mcp": {
"type": "http",
"url": "http://localhost:8080/mcp",
"env": {
"OPENAI_API_KEY": "sk-…",
"COROOT_API_URL": "https://coroot.your-company.com",
"COROOT_DEFAULT_PROJECT_ID": "production"
}
}
}
}
~/.claude.json)Claude CLI and Claude Desktop both read MCP servers from JSON config files. A minimal global configuration looks like:
{
"mcpServers": {
"coroot-mcp": {
"type": "http",
"url": "http://localhost:8080/mcp"
}
}
}
If you prefer to scope coroot-mcp to a single folder, add the same mcpServers block under the project entry in ~/.claude.json or in a per-project settings file (see Claude docs for the latest supported locations).
The MCP server exposes tools via tools/list and tools/call.
All tools are read-only and safe to expose to assistants by default.
OPENAI_API_KEY, COROOT_API_URL, COROOT_DEFAULT_PROJECT_ID).Environment variables (12-factor style):
OPENAI_API_KEY – API key used by Spring AI to talk to the OpenAI-compatible model.OPENAI_MODEL – optional, defaults to gpt-4.1-mini.COROOT_API_URL – base URL of your Coroot instance, defaults to https://coroot.vitayou.io.COROOT_DEFAULT_PROJECT_ID – default Coroot project ID when a tool call omits it (e.g. production).MCP_AUTH_TOKEN – optional bearer token required on /mcp when set. Clients must send Authorization: Bearer <token>.These map to Spring Boot configuration in src/main/resources/application.properties.
export OPENAI_API_KEY=sk-...
export MCP_AUTH_TOKEN=some-secret-token # optional
export COROOT_API_URL=https://coroot.your-company.com
export COROOT_DEFAULT_PROJECT_ID=production
./mvnw spring-boot:run
The MCP JSON-RPC endpoint will be available at:
POST http://localhost:8080/mcpIf MCP_AUTH_TOKEN is set, clients must send:
Authorization: Bearer <token>Once an image is published to a registry such as GitHub Container Registry, you can run it directly. For example, assuming:
ghcr.io/meritocra/coroot-mcp:0.2.0Run against a real Coroot instance:
docker run --rm -p 8080:8080 \
-e OPENAI_API_KEY=sk-... \
-e MCP_AUTH_TOKEN=some-secret-token \
-e COROOT_API_URL=https://coroot.your-company.com \
-e COROOT_DEFAULT_PROJECT_ID=production \
ghcr.io/meritocra/coroot-mcp:0.2.0
Run in stub mode (no Coroot required):
docker run --rm -p 8080:8080 \
-e MCP_AUTH_TOKEN=some-secret-token \
-e SPRING_PROFILES_ACTIVE=stub-coroot \
ghcr.io/meritocra/coroot-mcp:0.2.0
Build a container image using the provided multi-stage Dockerfile:
docker build -t coroot-mcp:latest .
Run it:
docker run --rm -p 8080:8080 \
-e OPENAI_API_KEY=sk-... \
-e COROOT_API_URL=https://coroot.your-company.com \
-e COROOT_DEFAULT_PROJECT_ID=production \
coroot-mcp:latest
It is recommended to set JAVA_OPTS for resource limits, for example:
docker run --rm -p 8080:8080 \
-e JAVA_OPTS="-Xms256m -Xmx512m" \
-e OPENAI_API_KEY=sk-... \
-e COROOT_API_URL=https://coroot.your-company.com \
-e COROOT_DEFAULT_PROJECT_ID=production \
coroot-mcp:latest
For local experiments, you can also use the provided docker-compose.yml:
docker compose up --build
By default this starts coroot-mcp with the stub-coroot profile and exposes the MCP endpoint on:
POST http://localhost:8080/mcpThe MCP manifest is defined in mcp.json. It declares:
http://localhost:8080/mcp.OPENAI_API_KEY, COROOT_API_KEY.OPENAI_MODEL, COROOT_API_URL, COROOT_DEFAULT_PROJECT_ID.MCP-compatible clients can either:
mcp.json in a project directory), orhttp://localhost:8080/mcp).Toolhive is a multi-server MCP manager that can run in your cluster or on the desktop.
mcp.json is accessible to Toolhive (either via local file path or by copying its contents into Toolhive’s MCP configuration).mcp.jsonhttp://localhost:8080/mcpinc-1 and suggest next steps.”Toolhive will call the MCP tools under the hood and present their outputs.
coroot-mcp) is licensed under the MIT License (see LICENSE).Community MCP server that exposes Coroot incidents and health summaries as read-only tools for AI assistants (not an official Coroot product).
工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
Coroot 透過 stdio 在本機執行。
上次檢查時間:2026-10-28。 Community MCP server that exposes Coroot incidents and health summaries as read-only tools for AI assistants (not an official Coroot product).