Beta. Tool names, capabilities, and install steps may still change. Not yet recommended for production.
A Model Context Protocol server that exposes the Epinio API as tools for AI agents such as Claude. It runs on your cluster and translates MCP tool calls into Epinio REST API requests, so an agent can deploy and manage applications through conversation.
By default the server wires only to the Epinio API, running as the calling
user — app lifecycle, logs, source retrieval, and CRUD for app charts, builder
images, and catalog services. A single elevated capability set that reaches
directly into Kubernetes — workload adoption — is off by default and
opt-in via the EPINIO_MCP_ELEVATED flag. See the reference docs.
Requires Epinio 1.14.1 or later. The server depends on the builder-image, catalog-service, and app-chart CRUD API and the source-retrieval endpoint, all introduced in Epinio 1.14.1; it will not work against earlier releases.
Full documentation lives at docs.epinio.io:
- Install the MCP server — prerequisites, configuration, and deployment.
- MCP server reference — the full tool list, optional capabilities, and health probes.
Clone the repo, set your cluster details in epinio.yml, then:
make setup # push the MCP to Epinio and smoke-test itThat is the core install — a pure Epinio-API server. To turn on the opt-in
elevated tier (workload adoption, which reaches directly into Kubernetes),
edit epinio-elevated.yml and run make elevated-setup instead.
See the install guide for
prerequisites and configuration, and make help for all targets.
make build # build dist/epinio-mcp with version stamped in
make test # unit tests with the race detector
make check # fmt-check + vet + lint + test (the CI gate)
make help # list all targetsTo run the server against a cluster locally:
EPINIO_API_URL=https://epinio.example.com EPINIO_USERNAME=admin EPINIO_PASSWORD=secret make runApache-2.0 — see LICENSE.