MCP Fetch

Instructions

Implementation Reference

• src/tools/puppeteer.ts:97-140 (handler)

  The main execution handler for the 'puppeteer' tool. It uses a switch statement on the action type to delegate to specific Puppeteer utility functions.

  export default async function puppeteer({ action }: InferSchema<typeof schema>) {
    try {
      switch (action.type) {
        // Browser actions
        case "list-browsers":
          return await listBrowsers()
  
        case "launch-browser":
          return await launchBrowser(action.headless, action.width, action.height, action.url)
  
        case "close-browser":
          return await closeBrowser(action.browserId)
  
        // Page actions
        case "list-pages":
          return await listPages()
  
        case "open-page":
          return await openPage(action.browserId, action.url)
  
        case "close-page":
          return await closePage(action.pageId)
  
        // Exec action
        case "exec-page":
          return await execPage(action.pageId, action.source)
  
        // Screenshot action
        case "take-screenshot":
          return await takeScreenshot(action.pageId, action.fullPage, action.format, action.quality)
  
        default:
          return JSON.stringify({
            success: false,
            error: "Invalid action type",
          }, null, 2)
      }
    } catch (error) {
      return JSON.stringify({
        success: false,
        error: error instanceof Error ? error.message : "Unknown error occurred",
      }, null, 2)
    }
  }

• src/tools/puppeteer.ts:13-82 (schema)

  Zod schema defining the input parameters for the puppeteer tool using a discriminated union based on 'action.type'.

  export const schema = {
    action: z.discriminatedUnion("type", [
      // Browser actions
      z.object({
        type: z.literal("list-browsers"),
      }),
      z.object({
        type: z.literal("launch-browser"),
        headless: z.boolean()
          .optional()
          .default(false)
          .describe("Whether to run browser in headless mode. Defaults to false"),
        width: z.number()
          .optional()
          .default(1280)
          .describe("Browser window width in pixels. Defaults to 1280"),
        height: z.number()
          .optional()
          .default(720)
          .describe("Browser window height in pixels. Defaults to 720"),
        url: z.string()
          .optional()
          .describe("Optional URL to navigate to after launching the browser"),
      }),
      z.object({
        type: z.literal("close-browser"),
        browserId: z.string().describe("The ID of the browser instance to close"),
      }),
      // Page actions
      z.object({
        type: z.literal("list-pages"),
      }),
      z.object({
        type: z.literal("open-page"),
        browserId: z.string().describe("The ID of the browser instance to create a page in"),
        url: z.string().optional().describe("Optional URL to navigate to after creating the page"),
      }),
      z.object({
        type: z.literal("close-page"),
        pageId: z.string().describe("The ID of the page to close"),
      }),
      // Exec action
      z.object({
        type: z.literal("exec-page"),
        pageId: z.string()
          .describe("The ID of the page to execute code on"),
        source: z.string()
          .describe("JavaScript code to execute. Will be executed outside of the page context. You can run commands like await page.goto('https://example.com') or await page.evaluate(() => { ... }). The code you write should return a string value that will serve as the result of the tool call."),
      }),
      // Screenshot action
      z.object({
        type: z.literal("take-screenshot"),
        pageId: z.string()
          .describe("The ID of the page to take a screenshot of"),
        fullPage: z.boolean()
          .optional()
          .default(false)
          .describe("Whether to take a screenshot of the full scrollable page. Defaults to false"),
        format: z.enum(["png", "jpeg"])
          .optional()
          .default("png")
          .describe("Image format for the screenshot. Defaults to png"),
        quality: z.number()
          .min(0)
          .max(100)
          .optional()
          .describe("Quality of the screenshot, only applicable for jpeg format (0-100)"),
      }),
    ]).describe("The action to perform with Puppeteer"),
  }

• src/tools/puppeteer.ts:85-94 (registration)

  Tool metadata that registers the tool with name 'puppeteer', description, and annotations for the MCP framework.

  export const metadata: ToolMetadata = {
    name: "puppeteer",
    description: "Control browsers and pages with Puppeteer - launch/close browsers, open/close pages, execute JavaScript, take screenshots, and more",
    annotations: {
      title: "Puppeteer Control",
      readOnlyHint: false,
      destructiveHint: false,
      idempotentHint: false,
    },
  }