Developer platform
Gessa developer resources
Build against Gessa’s supported generation API or authenticated remote MCP surface. This page is the stable index for machine contracts, authentication, failure handling, rate limits, and lifecycle expectations. Internal implementation routes are not a public compatibility promise.
Supported OpenAPI contract
The canonical machine-readable contract is https://api.gessa.ai/openapi.json. It describes the supported public generation surface. Generate clients and function-call tools from that contract, not from undocumented implementation routes.
The core workflow is: inspect capabilities, create a generation job, observe or fetch it, and cancel it when needed. Streaming event endpoints are documented separately from JSON function-call operations.
Authentication and safeguards
REST callers use a scoped workspace API key or short-lived capability token issued by Gessa. Send the credential only to https://api.gessa.ai. Start with read operations, inspect capabilities before generation, use idempotency where the contract exposes it, and require human approval before high-impact writes.
Agentic resource discovery
Domain-level clients can start with Gessa’s standards-aligned AI Catalog at gessa.ai/.well-known/ai-catalog.json. It advertises the public MCP Server Card and its HTTPS-bound identity metadata.
The Server Card at gessa.ai/.well-known/mcp-server-card.json follows the current draft MCP discovery extension. It declares identity and connection details only; clients still enumerate tools through MCP after authentication.
Remote MCP
The Streamable HTTP endpoint is https://api.gessa.ai/mcp/rpc. Remote clients use the OAuth 2.1 discovery metadata published under /.well-known/. A GET request to the RPC route may return 405 by design; MCP clients initialize with the protocol’s supported HTTP method and negotiated session headers.
Machine clients can discover the live remote transport from the MCP Registry manifest at gessa.ai/.well-known/mcp. The manifest also advertises the verified @gessa/mcp-server stdio proxy for local MCP hosts. Version 1.1.0 exposes the gessa-mcpexecutable and forwards to the same authenticated remote service.
Typed errors
Treat documented 4xx responses as requests for correction and 5xx/503 responses as potentially transient. Log the request identifier, preserve the machine-readable error code, and surface the human-readable message. Never assume an undocumented response shape or retry a mutation merely because a transport failed.
Rate limits and retries
Applicable API responses expose RateLimit and RateLimit-Policy. Slow down before the remaining budget reaches zero. On HTTP 429, wait forRetry-After; use bounded exponential backoff only for retry-safe operations.
Versioning and deprecation policy
The supported public contract follows semantic versions. Omit x-api-versionfor the current default or send x-api-version: 1 to pin major version 1. Unsupported major values fail explicitly. Additive compatible changes may ship within v1; breaking request or response changes require a new major API version. Gessa documents migrations and targets at least 90 days of notice before retiring a generally available version. Actually deprecated routes emit the standardDeprecation signal, a deprecation link, and a Sunset date when one has been scheduled. Current routes are never marked deprecated pre-emptively.
Read the dedicated, stable policy at gessa.ai/developers/api-lifecycle/.
Support and agent-readable indexes
See llms.txt, the agent catalog, and the Agent Skills index, and Gessa for AI assistants. Report contract or documentation problems to [email protected]. Integration and account support: [email protected].