@kubb/plugin-mcp 5.1.0 → 5.1.2

package/docs/overview.md

@@ -0,0 +1,44 @@
+# MCP
+
+The MCP plugin turns each operation into a typed MCP tool that AI assistants can call. Each tool uses Zod schemas from `@kubb/plugin-zod` for its input and output.
+
+Each tool handler calls the function a client plugin generates for that operation. The plugin writes `server.ts`, `start.ts`, and `.mcp.json` at the root of its `output.path`. `start.ts` calls the `startServer()` that `server.ts` exports, and `.mcp.json` runs `start.ts` to start a stdio server.
+
+## Setup
+
+- Add `@kubb/plugin-ts`, `@kubb/plugin-zod`, and a client plugin: `@kubb/plugin-fetch` or `@kubb/plugin-axios`. Leave `sdk` unset on the client plugin.
+- Install `@modelcontextprotocol/server` v2 and `zod` 4.2 or a later v4 release in the app that runs the server. Add `axios` too when you use `@kubb/plugin-axios`.
+- Install a TypeScript runner such as `tsx`. The generated `.mcp.json` runs `npx tsx`, which downloads `tsx` from npm when it is missing.
+- Copy the server entry from the generated `.mcp.json` into the `.mcp.json` your MCP client reads, such as the one at the project root for Claude Code.
+
+## Options worth knowing
+
+- `client` picks `fetch` or `axios` when both client plugins are registered. A single client is detected automatically.
+- `output.path` must be a folder name, such as the default `mcp`, because single-file output has no place for `server.ts`, `start.ts`, and `.mcp.json`.
+
+The `args` path in `.mcp.json` points at `start.ts` relative to the Kubb `root`, because MCP clients start the command from their own working directory. Use an absolute path for a client that starts in another folder, or point it at your own entry file to run setup code before `startServer()`.
+
+See [Options](https://kubb.dev/extensions/kubb-labs/plugin-mcp/options) for every option and its default.
+
+## Example
+
+The config below generates an MCP server whose tools call the Petstore API through `@kubb/plugin-fetch`.
+
+```typescript [kubb.config.ts]
+import { defineConfig } from 'kubb/config'
+import { pluginTs } from '@kubb/plugin-ts'
+import { pluginZod } from '@kubb/plugin-zod'
+import { pluginFetch } from '@kubb/plugin-fetch'
+import { pluginMcp } from '@kubb/plugin-mcp'
+
+export default defineConfig({
+  input: './petStore.yaml',
+  output: { path: './src/gen' },
+  plugins: [
+    pluginTs(),
+    pluginZod(),
+    pluginFetch({ baseURL: 'https://petstore.swagger.io/v2' }),
+    pluginMcp({ output: { path: 'mcp', mode: 'directory' } }),
+  ],
+})
+```

package/kubb-extension.json

@@ -11,9 +11,16 @@
     "dark": "https://kubb.dev/feature/mcp-dark.svg"
   },
   "tags": ["mcp", "model-context-protocol", "ai", "claude", "llm", "codegen", "openapi"],
-  "docs": "https://kubb.dev/plugins/plugin-mcp",
+  "docs": "https://kubb.dev/extensions/kubb-labs/plugin-mcp",
+  "maintainers": [
+    {
+      "name": "Stijn Van Hulle",
+      "github": "stijnvanhulle"
+    }
+  ],
+  "dependencies": ["plugin-ts", "plugin-zod"],
   "compatibility": {
-    "kubb": ">=5.8.0",
+    "kubb": ">=5.8.3",
     "node": ">=22"
   },
   "role": "consumer",
@@ -90,7 +97,7 @@
       "key": "client",
       "path": ["client"],
       "label": "Client",
-      "description": "HTTP client the generated hooks call. Only needed to disambiguate when both plugin-fetch and plugin-axios are enabled",
+      "description": "HTTP client the generated handlers call. Only needed to disambiguate when both plugin-fetch and plugin-axios are enabled",
       "type": "select",
       "default": null,
       "choices": [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@kubb/plugin-mcp",
-  "version": "5.1.0",
+  "version": "5.1.2",
   "description": "Generate Model Context Protocol (MCP) tool definitions with Kubb. Expose your REST APIs as AI-callable tools for LLMs, Claude, ChatGPT, and other AI assistants.",
   "keywords": [
     "ai",
@@ -14,7 +14,7 @@
     "model-context-protocol",
     "typescript"
   ],
-  "homepage": "https://kubb.dev/plugins/plugin-mcp",
+  "homepage": "https://kubb.dev/extensions/kubb-labs/plugin-mcp",
   "license": "MIT",
   "author": "stijnvanhulle",
   "repository": {
@@ -39,7 +39,7 @@
   "files": [
     "dist",
     "kubb-extension.json",
-    "OVERVIEW.md",
+    "docs",
     "!/**/**.test.**",
     "!/**/__tests__/**",
     "!/**/__snapshots__/**"
@@ -63,20 +63,20 @@
     "registry": "https://registry.npmjs.org/"
   },
   "dependencies": {
-    "@kubb/plugin-ts": "^5.0.8",
-    "@kubb/plugin-zod": "^5.4.6"
+    "@kubb/plugin-ts": "^5.0.10",
+    "@kubb/plugin-zod": "^5.4.8"
   },
   "devDependencies": {
     "@internals/client": "0.0.0",
     "@internals/shared": "0.0.0",
     "@internals/utils": "0.0.0",
     "@modelcontextprotocol/server": "^2.3.1",
-    "kubb": "^5.8.1",
+    "kubb": "^5.8.3",
     "zod": "^4.6.5"
   },
   "peerDependencies": {
     "@modelcontextprotocol/server": "^2.0.0",
-    "kubb": "^5.8.1",
+    "kubb": "^5.8.3",
     "zod": "^4.2.0"
   },
   "peerDependenciesMeta": {

package/OVERVIEW.md

@@ -1,15 +0,0 @@
-# MCP
-
-Generates a Model Context Protocol (MCP) server from your OpenAPI spec. Each operation becomes a typed tool that AI assistants can call.
-
-The tool handlers call a function generated by a client plugin, and Zod schemas from `@kubb/plugin-zod` validate the tool input. The plugin writes `server.ts` and `.mcp.json` next to the handlers.
-
-## Setup
-
-- Add `@kubb/plugin-ts`, `@kubb/plugin-zod`, and a client plugin: `@kubb/plugin-fetch` or `@kubb/plugin-axios`.
-- Install `@modelcontextprotocol/server` v2 or higher and `zod` 4.2 or higher in the app that runs the server.
-
-## Options worth knowing
-
-- `client` picks `fetch` or `axios` when both client plugins are registered. A single client is detected automatically.
-- Keep `output.path` a folder name, such as the default `mcp`. Single-file output has no place for `server.ts` and `.mcp.json`.