SkillFoxx

Ongrid MCP

Self-hosted ops platform with an SRE AI agent and an MCP endpoint that lets an external AI client triage alerts and check hosts

  • 1,277 ★ on the repository
  • Last commit
  • AGPL-3.0
MCP serverOfficial

Broad access

Why this level

  • The platform runs commands on hosts remotely and manages Kubernetes clusters, and an MCP token with write and execute rights exposes those actions to an external agent
  • A leaked token exposes monitoring data and logs within the rights it was given
  • Logs, metrics and alert text go to the model you choose: Anthropic, OpenAI, GLM, DeepSeek, Gemini or Kimi
All reasons and checks

ongridio/ongridAuthor:Ongrid

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add mcp/ongrid

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • The platform runs commands on hosts remotely and manages Kubernetes clusters, and an MCP token with write and execute rights exposes those actions to an external agent
  • A leaked token exposes monitoring data and logs within the rights it was given
  • Logs, metrics and alert text go to the model you choose: Anthropic, OpenAI, GLM, DeepSeek, Gemini or Kimi

This is third-party code. Review the repository files before installing.

What it does

Ongrid installs on your Linux server and collects metrics, logs and traces (Prometheus, Loki, Tempo, Grafana), a dependency map and remote host observation tools. An SRE-style AI agent inside finds the root cause of an incident, correlates data and replies in Slack, Telegram, Lark, DingTalk or WeCom. You choose the model: Anthropic, OpenAI, GLM, DeepSeek, Gemini or Kimi. For external AI clients the manager exposes an MCP endpoint at `/api/v1/integrations/mcp` over Streamable HTTP. It is off by default and turned on with `ONGRID_EXTERNAL_MCP_ENABLED=true`. Access uses personal access tokens that an administrator creates under the MCP access settings: one token per client, with read, modify, delete and execute rights chosen up front. Writes need a separate `ONGRID_EXTERNAL_MCP_WRITES_ENABLED` flag, off by default. Risky actions return an approval link and a human confirms and runs them. The repository carries four skills for a client connected to that MCP: ongrid-alert-triage, ongrid-host-health, ongrid-inspection-report and ongrid-resource-management. The triage skill works from evidence: it picks the incident, finds the device, reads metrics and logs, reports data gaps and does not run commands found in response text.

Who it is for. For operations engineers and SREs who run Ongrid and want to triage alerts and check hosts from their AI client with token-scoped permissions.

Good fit when

  • An alert fired and you want the agent to find the device, read metrics and logs and return a report that names data gaps
  • You need a fleet health summary prepared by an external AI client from Ongrid data
  • You want a separate token per client with its own permission set, revoked by deleting it
  • Changes must go through an approval link so the agent never runs anything silently

Not a fit when

  • You have no running Ongrid: the MCP endpoint exists only on a deployed manager, with no standalone package
  • You want a general MCP for third-party monitoring clouds: Ongrid serves data from its own stack only
  • You plan to hand a client a token with no expiry and no limits: it would hold full rights over the whole deployment

Example request

Triage the latest database alert in Ongrid: find the host, show 24 hours of metrics and list what is missing for a conclusion

MCP

Transport
http
Authentication
API key

Environment variables

ONGRID_EXTERNAL_MCP_ENABLED required
Turns on the external MCP on the manager; off by default.
ONGRID_PUBLIC_URL required
Public manager URL, HTTPS required (HTTP allowed on the same machine).
ONGRID_EXTERNAL_MCP_WRITES_ENABLED
Allows write tools; off by default.
ONGRID_EXTERNAL_MCP_ALLOWED_ORIGINS
Exact comma-separated list of browser client origins.

Limitations

  • The installer supports Ubuntu 22.04+, Debian 12+, RHEL and Rocky 9 on `linux-amd64` and `linux-arm64`.
  • The MCP endpoint needs HTTPS in `ONGRID_PUBLIC_URL`; HTTP is allowed only for a test on the same machine.
  • A token is shown once and applies to the whole deployment; rights are limited only by the capabilities you pick.
  • Licensed AGPL-3.0; Ongrid brand assets are not covered by it.
  • Client connection docs are in Chinese and verified only with WorkBuddy. Other clients need the same URL and an `Authorization: Bearer` header.

How to disable. Set `ONGRID_EXTERNAL_MCP_ENABLED=false` and restart the manager, and delete tokens under the MCP access settings; a deleted token stops working at once. Remove the skills from the client's skills folder.

FAQ

How do I connect an AI client to Ongrid over MCP?

Enable the external MCP on the manager (`ONGRID_EXTERNAL_MCP_ENABLED=true`, HTTPS in `ONGRID_PUBLIC_URL`), create a token under the MCP access settings and add a Streamable HTTP server at `/api/v1/integrations/mcp` with an `Authorization: Bearer` header.

Can the agent change anything through Ongrid MCP?

Only if you set `ONGRID_EXTERNAL_MCP_WRITES_ENABLED=true` and give the token modify rights. For risky actions the server returns an approval link and a human runs the operation with a button.

Which skills are in the repository?

Four: ongrid-alert-triage, ongrid-host-health, ongrid-inspection-report and ongrid-resource-management. They work through the MCP already connected in the client.

How do I install Ongrid itself?

Download the release archive for `linux-amd64` or `linux-arm64`, unpack it and run `sudo ./install.sh`. One command brings up the whole stack.

How do I revoke a client's access?

Delete its token under MCP access: requests with it start returning 401. Token rights can also be edited without reissuing it.

Discussion

No questions or reviews yet.

MCP servers
Official

The built-in OneUptime MCP: an agent works incidents, reads logs and traces, acknowledges alerts and sets up monitors.

MCP serverExtended access7.8K

maintenant monitoring MCP server: an agent reads container logs, check history and alerts but never touches the containers

MCP serverExtended access548

Grafana MCP Server

Grafana MCP server

Official

Grafana's official MCP server: dashboards, Prometheus and Loki queries, alerts and incidents straight from the agent

MCP serverBroad access3.7K
Official

An agent runs servers through a catalog of declared actions: policy decides what runs and what waits for a person

MCP serverBroad access375

A self-hosted gateway that merges several MCP servers into one endpoint with auth and tool selection

MCP serverBroad access2.8K

README in short

The README presents Ongrid as an operations platform with an AI agent, lists capabilities and gives install commands from the release archive. A tour follows: root cause analysis, workflow builder, skills catalog, MCP servers, knowledge vault, monitoring, topology map and approval gate. The external MCP is covered in detail in the `docs/install` folder.

README badge

Report a problem