browser-pool-mcp

Instructions

Implementation Reference

• index.js:424-432 (registration)

  Registration of the browser_wait_for tool in the MCP server. It delegates the execution to proxyToolCall.

  server.tool('browser_wait_for', 'Wait for condition', {
    time: z.number().optional(),
    text: z.string().optional(),
    textGone: z.string().optional()
  }, async (args) => {
    const check = requireActivePage();
    if (check) return check;
    return proxyToolCall('browser_wait_for', args);
  });

• index.js:209-237 (helper)

  The proxyToolCall function acts as a wrapper that forwards tool execution requests to an underlying MCP client instance.

  async function proxyToolCall(toolName, args) {
    log(`[proxyToolCall] ${toolName} with args: ${JSON.stringify(args)}`);
    const { client } = await getOrCreateInstance();
    log(`[proxyToolCall] got client for port ${assignedPort}`);
  
    // Update last used
    if (assignedPort && instances.has(assignedPort)) {
      instances.get(assignedPort).lastUsed = Date.now();
    }
  
    try {
      log(`[proxyToolCall] Calling client.callTool...`);
      const result = await client.callTool({ name: toolName, arguments: args || {} });
      log(`[proxyToolCall] Result type: ${typeof result}`);
      log(`[proxyToolCall] Result: ${JSON.stringify(result).slice(0, 500)}`);
  
      // The SDK returns { content: [...], isError?: boolean }
      // We need to return this same format
      if (result && result.content) {
        return result;
      }
  
      // Fallback: wrap in content array if needed
      return {
        content: [{ type: 'text', text: JSON.stringify(result) }]
      };
    } catch (error) {
      log(`[proxyToolCall] ERROR: ${error.message}\n${error.stack}`);
      return {