RATATOSKRATATOSK
Sign in

MCP server

MCP (Model Context Protocol) is a standard for giving AI agents tools. With Ratatosk's MCP server attached, an agent can query release facts directly and check them against a running stack.

  • Source, install, configuration (canonical): github.com/garlicKim21/ratatosk-mcp
  • Container image: ghcr.io/garlickim21/ratatosk-mcp
  • Read-only. It changes nothing in your cluster.

The four tools

Tool What it does
list_facts Incremental fact feed (project, type, severity filters; cursor polling)
facts_by_entity Every fact touching one identifier (CVE, CRD, flag, …)
get_release One full release review, optionally with the original note
check_stack Compares your running versions against the facts and briefs the upgrade path

The privacy design of check_stack

Running versions can be sensitive. check_stack performs the version comparison inside the MCP server process. Only project slugs are sent to ratatosk.io; version numbers never leave the process. The public repository lets you verify this in code.

YOUR INFRASTRUCTUREAI agentClaude Code · kagent① Stack, versions includedenvoy v1.36.8⑤ Briefingbriefingratatosk-mcpruns on your side④ Local version compare② Project names onlyenvoy③ release factsfacts ×51ratatosk.iofacts database⑥ Briefing received · next stepcritical 1 · 3 action items→ upgrade envoy to v1.36.9Detailed version info never leaves your infrastructure

Ways to run it

  • stdio: for local agents such as Claude Code. A single binary is enough.
  • HTTP (in-cluster): set MCP_HTTP_ADDR and it serves streaming HTTP, so agents inside the cluster connect through a Service. A Helm chart ships in the repository.

Install commands and concrete configuration examples are in the repository README (English, Korean, Japanese).

Server card

For machine discovery, a standard server card is published at /.well-known/mcp/server-card.json.