Evidence-first authority page
How to Wrap a REST API into an MCP Server
A practical guide to wrapping an existing internal REST API as an MCP server, so an agent can call it as a tool.
Direct answer
To wrap a REST API as an MCP server, define one tool per meaningful API operation (not a single generic 'call any endpoint' tool), map each endpoint's parameters to a strict input schema, handle the HTTP call and error translation inside the tool handler, and return structured results the model can reason about directly.
Last reviewed 2026-08-29
This page remains noindex while its page-level source receipts are completed and verified. It is available for review, not presented as publication-ready authority.
Why one tool per operation, not a generic passthrough
A single 'call this API with any path and method' tool is easier to build but gives the model far more room to construct an invalid or unintended request, and gives you no natural place to apply per-operation validation or permission checks. Defining a distinct tool per meaningful operation (get_order_status, create_ticket) is more setup work but produces a materially more reliable and secure integration.
Practical steps
- Identify the specific API operations the agent's task actually needs — don't wrap the entire API surface if only a few endpoints matter.
- Write a strict JSON schema for each tool's inputs, validated before the HTTP call is made.
- Translate API errors into a clear message returned to the model, rather than letting a raw HTTP error or stack trace pass through.
- Consider rate limits on the underlying API and apply them at the wrapper level so a fast agent loop can't exceed them.