Skip to main content
  • Pricing

Developer hub

Oakgen Developer Hub

Oakgen exposes one live developer interface: a remote Streamable HTTP Model Context Protocol (MCP) endpoint for compatible MCP clients. It lets an authenticated account inspect current models and schemas, get a quote, upload references, start work, monitor the result, and request cancellation.

Endpoint and authentication

Connect your MCP client to https://oakgen.ai/api/mcp. The endpoint uses HTTP bearer authentication and supports OAuth discovery through /.well-known/oauth-protected-resource/mcp and /.well-known/oauth-authorization-server when OAuth is enabled for the deployment.

You can also create a Personal API key in the MCP console and use it as a bearer token. Personal keys can be limited to models:read, generations:read, and generations:write. OAuth connections currently receive the full generation authority and do not offer selectable scopes.

Current MCP tools

search_models
Find approved image, video, and audio models.
get_model_schema
Read the exact public input schema for a model.
get_pricing
Validate model input and receive an exact generation quote.
upload_media
Create or complete an owner-scoped upload for local media.
start_generation
Start an idempotent generation and receive its durable handle.
get_generation_status
Read authoritative owner-scoped generation state and assets.
cancel_generation
Request cancellation and receive the current generation state.

Safe generation workflow

  1. Use search_models and get_model_schema to select a model and construct valid input.
  2. Use get_pricing with the intended input. Request a quote before starting paid generation.
  3. Use upload_media for a local image, video, or audio reference; do not send base64 or data URLs.
  4. Call start_generation with the approved input, a maxCredits ceiling, and a stable idempotencyKey. Reuse that key only for retries of the same request.
  5. Poll get_generation_status with the returned generation ID, or call cancel_generation when cancellation is required.

Limits and errors

The current policy permits 60 requests per minute and 1,000 requests per day for an MCP identity. Generation starts are additionally limited to 10 per minute and 1,000 per day. A limited request returns HTTP 429 with retry information; unavailable authentication authority returns HTTP 503. Invalid JSON-RPC requests and unknown tools are reported with JSON-RPC error responses, while authorization and connection failures return their applicable HTTP status.

Model availability, schema, price, accepted references, and output shape are current at the time of a request. Inspect the tools rather than assuming a fixed catalog. MCP usage and web usage share the same Oakgen account and credit balance.

Developer FAQ

What is the Oakgen MCP endpoint?

It is a remote Streamable HTTP Model Context Protocol endpoint at https://oakgen.ai/api/mcp. Compatible MCP clients use it to inspect models and schemas, get a quote, upload references, start generations, check status, and request cancellation.

How do I authenticate MCP requests?

The endpoint uses HTTP bearer authentication. Use OAuth discovery where it is enabled, or create a Personal API key in the MCP console. Personal keys can be limited to models:read, generations:read, and generations:write.

How much does an MCP generation cost?

Call get_pricing with the intended input to get an exact quote, then pass a maxCredits ceiling to start_generation. MCP and web usage share the same Oakgen account and credit balance.

What are the MCP rate limits?

An MCP identity can make 60 requests per minute and 1,000 per day, with generation starts limited to 10 per minute and 1,000 per day. A limited request returns HTTP 429 with retry information.

Need account help? Contact [email protected].