Do I need to rewrite my backend API to make it usable by AI agents?

No. If your REST API has a well-maintained OpenAPI specification, tools such as Speakeasy, FastMCP and openapi-mcp-generator can read that spec and convert each endpoint into a tool definition for agents, without changing the backend itself.

What still needs human work after auto-generating MCP tools from OpenAPI?

Authorization and error handling, consolidating multiple endpoints into fewer tools, and refining the descriptions so agents can judge what each tool does and when to use it.

Why can't an agent just use every endpoint as its own tool?

Too many tools strain the context and waste tokens; turning an OpenAPI spec wholesale into tools would make 200 endpoints into 200 tools, so operations need to be grouped behind a smaller number of tools.

Large Language ModelsAI Coding AssistantsITmedia AI＋40m agoNov 12, 2026, 06:01 JST

MCP now joins the Web, but agents still need OpenAPI specs and human-tuned tools

Send on LINE

3 Key Points

What happened
In a Postman column, Akihiko Kusanagi says MCP's July 2026 revision made MCP servers stateless, ordinary HTTP parts, so tools like Speakeasy, FastMCP and openapi-mcp-generator can generate MCP servers from existing OpenAPI specs without rewriting backend APIs.
Why it matters
That shifts API design from machine-readable specs toward machine-interpretable ones, where agents decide which operation to call, so descriptions must explain what each tool does and when to use it.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Get these free every morning

Context & Analysis

The column builds on an earlier installment's point that MCP's July 2026 revision made MCP servers stateless, which the author reframes as MCP ceasing to be a special construct and becoming an ordinary HTTP component that joins the API world. From there, the practical path for enterprises is to start from an existing OpenAPI specification: tools such as Speakeasy, FastMCP and openapi-mcp-generator read that spec and convert endpoints into agent-facing tool definitions, so the backend API does not need to be rewritten.

The author draws a contrast between conventional machine processing and AI. SDK generation follows fixed rules, but an agent interprets the meaning of descriptions and decides which operation to use based on its goal and situation. Kevin Swiber's description of MCP as the "ultimate API consumer" captures this relationship. As a result, a specification that is merely machine-readable is not enough; the descriptions must be refined so an AI can judge what a tool can do and when it should be used.

The column also notes that automatic generation is a convenient starting point but, especially for large APIs, not necessarily production-ready. Authorization and error handling, consolidating multiple endpoints into tools, and crafting descriptions are left for people to handle after generation. Around MCP, the author mentions that MCP Apps was published on January 26, 2026 as the first official MCP extension, and that the July 28, 2026 revision positioned a formal framework for extensions within the specification; related approaches such as Google's A2UI and AG-UI are still being sorted out. The next installment is set to address agentic commerce, where agents call APIs, choose products and complete payments.

FAQ

Do I need to rewrite my backend API to make it usable by AI agents?: No. If your REST API has a well-maintained OpenAPI specification, tools such as Speakeasy, FastMCP and openapi-mcp-generator can read that spec and convert each endpoint into a tool definition for agents, without changing the backend itself.
What still needs human work after auto-generating MCP tools from OpenAPI?: Authorization and error handling, consolidating multiple endpoints into fewer tools, and refining the descriptions so agents can judge what each tool does and when to use it.
Why can't an agent just use every endpoint as its own tool?: Too many tools strain the context and waste tokens; turning an OpenAPI spec wholesale into tools would make 200 endpoints into 200 tools, so operations need to be grouped behind a smaller number of tools.

ITmedia AI＋Read Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.