mcp2cli-runtime-api-tooling

Run mcp2cli with the --spec flag pointing to a local or remote OpenAPI file, plus --base-url for the API endpoint. Use --list to see generated subcommands, then call endpoints directly, for example mcp2cli --spec ./openapi.yaml --base-url http://localhost:8000 list-pets --limit 20.

Use mcp2cli with --mcp for HTTP/SSE servers or --mcp-stdio for local servers launched via stdin/stdout. List tools with --list, then invoke a tool as a subcommand with its arguments, such as mcp2cli --mcp-stdio "npx @modelcontextprotocol/server-filesystem /tmp" read-file --path /tmp/data.json.

Yes, mcp2cli supports OAuth across all modes, including authorization code with PKCE and client credentials flows. Tokens are cached in ~/.cache/mcp2cli/oauth/ and refreshed automatically when expired.

Use the env: or file: prefixes so secrets are read from environment variables or files instead of appearing in process lists. For example, pass --auth-header "Authorization:Bearer env:API_TOKEN" to load the token from the API_TOKEN environment variable.

Specs with relative server definitions need an explicit --base-url flag so requests resolve correctly. Also note that local file specs are never cached, while remote specs cache for one hour by default and can be refreshed with --refresh.

Instead of sending full tool schemas (~1,400 tokens for 96 tools), agents use --list --compact for ~20 tokens of tool names. Usage-aware ranking with --top N and TOON output format further reduce tokens by 40-60% on large responses.