Rest Api to Mcp Wrapper

Your agent needs to talk to a REST API. Writing the MCP server by hand means translating every endpoint into a tool definition, getting the schemas right, handling pagination, and mapping errors — hours of careful, boring work per API. This skill automates it: give it an OpenAPI/Swagger spec (or even a plain endpoint list), and it generates a complete MCP server with one tool per endpoint, sensible input schemas derived from the spec, pagination helpers, and error mapping — plus a review pass that fixes the things generators always get wrong.

To convert a REST API contract into a working MCP server with minimal human effort: parse the spec, generate one well-described tool per operation, derive strict JSON schemas from request bodies and parameters, add pagination and auth wiring, and produce the same production hygiene as a hand-built server (structured errors, logging, tests). The human's job shrinks to reviewing the generated tool descriptions and approving the auth mapping.

• When you want your agent to use a third-party REST API (Stripe-style, internal platform API, SaaS API) and an OpenAPI spec exists.

• When the spec is missing but you have docs or an endpoint list — the skill works from those too, marking generated schemas as assumed where it had to guess.

• When you need to wrap several related APIs and want consistent tool naming and error shapes across all of them.

• When an API adds new endpoints and you need to regenerate without losing your hand-written customizations (the skill uses clearly-marked GENERATED vs CUSTOM regions).

• When the API is GraphQL, gRPC, or event-driven — the generation heuristics are REST/HTTP-specific and will produce nonsense.

• When you are designing an API from scratch — use mcp-server-scaffolder for greenfield tool design; this skill wraps what already exists.

• When the spec is so broken it cannot be parsed (the skill will tell you, and suggest fixing the spec first — possibly with api-contract-reviewer).

• For APIs requiring complex multi-step workflows per tool (e.g. "create order" needs 4 calls with state) — generated tools are one-call-per-tool; orchestration belongs in the agent, or hand-write those tools.