The Octri Blog
Product updates and practical writing on API documentation, SDKs, and developer experience.
Showing articles tagged #mcp.

Compare agent interfaces with the same task, permissions and outcome checks. Token counts mean little when one path does less work.

Follow identity from MCP client to downstream API. Check token audiences, delegated credentials and tenant authorization at each boundary.

Design bounded responses that preserve useful facts. Filter, select and aggregate before large intermediate payloads enter an agent's context.

A tool can avoid writes while returning hostile instructions. Treat customer-controlled API content as untrusted data throughout an agent workflow.

There is no spec format for MCP servers yet, so what an integrator gets is a README and a config snippet. Here is what actually has to be in it.

Exposing your whole API to an agent makes it worse, not more capable. Tool choice is a selection problem, and every extra option makes the selection harder.

The comparison is framed as a replacement and it is not one. One describes an API to people and compilers. The other hands a running agent a set of things it may do.

A tool call is a real request against a real API. Before you connect one, the question is not which endpoints to expose. It is which ones an agent must never reach.

An agent asked to integrate your API will recite one it saw in training or guess from the name. Both produce code that looks right and does not run. The fix is not a better prompt.