Cyborg MCP Servers

Model Context Protocol (MCP) servers for the Cyborg ecosystem, packaged as a single installable distribution with per-server extras. Usable standalone with Claude Code, Cursor, and other MCP clients, and programmatically in-process by other Python hosts.

Servers

Both servers' code ships in every install; the extras gate dependencies, not modules. A server whose extra is missing errors with a friendly hint (e.g. install 'cyborg-mcp-servers[argo]') instead of a raw ImportError.

Install

Requires Python 3.11+. Tagged releases publish to PyPI (via Trusted Publishing); you can also install from a GitHub Release asset or directly from git. (The first PyPI publish needs a one-time project + Trusted Publisher setup — see RELEASING.md.)

# Both servers, all transports and data sources:
uv pip install "cyborg-mcp-servers[all] @ git+https://github.com/openshift-eng/cyborg-mcp-servers"

# Just one server:
uv pip install "cyborg-mcp-servers[argo] @ git+https://github.com/openshift-eng/cyborg-mcp-servers"
uv pip install "cyborg-mcp-servers[orgdata,gcs] @ git+https://github.com/openshift-eng/cyborg-mcp-servers"

Extras: argo, orgdata, gcs (orgdata GCS data source), http (streamable-HTTP transport; also ships transitively with mcp), dev, and all (= argo,orgdata,gcs,http).

Run

argo-k8s-mcp                 # stdio (default); or: argo-k8s-mcp http --port 8080
orgdata-mcp-server stdio     # or: orgdata-mcp-server http --port 8000

Client configuration (.mcp.json)

{
  "mcpServers": {
    "argo-k8s": {
      "type": "stdio",
      "command": "argo-k8s-mcp",
      "env": { "ARGO_MCP_KUBECONFIG_CONTEXT": "your-cluster-context" }
    },
    "orgdata": {
      "type": "stdio",
      "command": "orgdata-mcp-server",
      "args": ["stdio"],
      "env": {
        "ORGDATA_GCS_BUCKET": "resolved-org",
        "ORGDATA_GCS_OBJECT_PATH": "orgdata/comprehensive_index_dump.json",
        "ORGDATA_GCS_PROJECT_ID": "openshift-crt"
      }
    }
  }
}

See each server's doc for the full configuration reference.

In-process use

argo-k8s exposes its tools as an importable library for agent frameworks — see argo_k8s_mcp.api and docs/argo-k8s.md. orgdata is best consumed in-process via its underlying library, orgdatacore, rather than the MCP server.

Development

make install-dev    # uv sync with all extras + dev tooling
make all            # lint + format-check + typecheck + test (the CI gate)
make test           # pytest (both suites, under tests/)

See CONTRIBUTING.md and RELEASING.md.

Layout

cyborg-mcp-servers/
├── pyproject.toml          # one package, per-server extras, hatch-vcs version
├── src/
│   ├── argo_k8s_mcp/       # argo-k8s server + in-process api
│   └── orgdata_mcp/        # orgdata server
├── tests/{argo_k8s,orgdata}/
├── deploy/argo-k8s/        # RBAC manifests for in-cluster deployment
├── scripts/                # cluster setup helpers
└── docs/                   # per-server references