OpenAPI spec and docs MCP

Every endpoint you can call with an API key — inference under /v1, and the Compute API for GPU instances, clusters, storage, SSH keys, container images and templates — is described in one machine-readable document. It is generated from the API itself, so it always matches what is running.

What	Where
OpenAPI spec (JSON)	https://api.ecohash.com/openapi.json
Browsable reference	https://api.ecohash.com/docs
Docs MCP server	https://mcp.ecohash.com/docs

None of them needs an API key to read.

Generate a client

The spec is OpenAPI 2.0 (Swagger). OpenAPI Generator reads it directly:

npx @openapitools/openapi-generator-cli generate \
  -i https://api.ecohash.com/openapi.json -g python -o ecohash-client

Tools that only accept OpenAPI 3 can convert it first:

npx swagger2openapi https://api.ecohash.com/openapi.json -o ecohash-openapi3.json

For inference you can also keep using any OpenAI-compatible SDK — see Chat completions.

Docs MCP server

Give a coding agent the API reference as tools, so it looks up an endpoint instead of guessing:

claude mcp add -s user --transport http ecohash-docs https://mcp.ecohash.com/docs

For other clients, add https://mcp.ecohash.com/docs as a remote (HTTP) MCP server with no headers.

Tool	What it does
list_endpoints	Every endpoint, or only one group (for example gpu-instances)
search_endpoints	Find endpoints by words — terminate, ssh key, embeddings
get_endpoint	Parameters, request body, responses and required access for one endpoint

The docs server only describes the API; it cannot call it and never sees your key. To launch and manage GPUs from an agent, use the GPU MCP server.