iframer-cli 1.0.2 → 1.0.4

package/cli.cjs

@@ -201,20 +201,6 @@ const [,, command, ...args] = process.argv;
 
 async function main() {
   switch (command) {
-    case "install": {
-      try {
-        execSync("claude mcp add iframer -- npx -y --package=iframer-cli iframer-mcp", { stdio: "inherit" });
-        console.log("\n  iframer MCP installed in Claude Code!");
-        console.log("  Now run: iframer login");
-      } catch {
-        console.error("  Failed to install. Make sure Claude Code CLI is available.");
-        console.error("  You can manually add it with:");
-        console.error("  claude mcp add iframer -- npx -y --package=iframer-cli iframer-mcp");
-        process.exit(1);
-      }
-      break;
-    }
-
     case "login":
       await login();
       break;
@@ -418,6 +404,111 @@ async function main() {
       break;
     }
 
+    // ─── Proxy tunnel ───────────────────────────────────────────────
+
+    case "proxy": {
+      const token = requireToken();
+      const server = getServer();
+      const wsUrl = server.replace(/^http/, "ws") + "/proxy/tunnel?token=" + token;
+
+      let WebSocket;
+      try {
+        WebSocket = require("ws");
+      } catch {
+        console.error("  ws package required. Run: npm install -g ws");
+        console.error("  Or: bun add ws");
+        process.exit(1);
+      }
+
+      const net = require("net");
+      const ws = new WebSocket(wsUrl);
+      const connections = new Map();
+
+      console.log("\n  Connecting to server...");
+
+      ws.on("open", () => {
+        console.log("  Connected! Your browser sessions will route through this machine.");
+        console.log("  Keep this running. Press Ctrl+C to stop.\n");
+      });
+
+      ws.on("message", (msg) => {
+        if (Buffer.isBuffer(msg) && msg.length > 4) {
+          // Binary data frame: first 4 bytes = connId
+          const connId = msg.readUInt32BE(0);
+          const data = msg.slice(4);
+          const socket = connections.get(connId);
+          if (socket && !socket.destroyed) {
+            socket.write(data);
+          }
+          return;
+        }
+
+        try {
+          const data = JSON.parse(msg.toString());
+
+          if (data.type === "ready") {
+            console.log(`  Proxy tunnel active (server port ${data.port})`);
+          }
+
+          if (data.type === "connect") {
+            const { id, host, port } = data;
+            const socket = net.createConnection({ host, port }, () => {
+              ws.send(JSON.stringify({ type: "connected", id }));
+            });
+
+            socket.on("data", (chunk) => {
+              const header = Buffer.alloc(4);
+              header.writeUInt32BE(id);
+              if (ws.readyState === 1) ws.send(Buffer.concat([header, chunk]));
+            });
+
+            socket.on("end", () => {
+              ws.send(JSON.stringify({ type: "close", id }));
+              connections.delete(id);
+            });
+
+            socket.on("error", (err) => {
+              ws.send(JSON.stringify({ type: "error", id, message: err.message }));
+              connections.delete(id);
+            });
+
+            connections.set(id, socket);
+          }
+
+          if (data.type === "close") {
+            const socket = connections.get(data.id);
+            if (socket) {
+              socket.destroy();
+              connections.delete(data.id);
+            }
+          }
+        } catch {}
+      });
+
+      ws.on("close", () => {
+        console.log("\n  Disconnected from server.");
+        process.exit(0);
+      });
+
+      ws.on("error", (err) => {
+        console.error(`  Connection error: ${err.message}`);
+        process.exit(1);
+      });
+
+      process.on("SIGINT", () => {
+        console.log("\n  Stopping proxy tunnel...");
+        for (const [, socket] of connections) {
+          try { socket.destroy(); } catch {}
+        }
+        ws.close();
+        process.exit(0);
+      });
+
+      // Block forever
+      await new Promise(() => {});
+      break;
+    }
+
     // ─── Screenshot ────────────────────────────────────────────────
 
     case "screenshot": {
@@ -532,7 +623,6 @@ async function main() {
   iframer - CLI for the Agentic Browser API
 
   Commands:
-    install                        Install iframer MCP in Claude Code
     login                          Authenticate with the server
     logout                         Remove saved credentials
     status                         Show current auth status

package/index.js

@@ -75,8 +75,13 @@ WORKFLOW:
 6. For logins: check credentials with "status". If credentials are missing, call "store_credentials" to prompt the user — NEVER tell them to use the CLI. Then use "login" tool.
 7. Always call "interactive_stop" when done to save session state.
 
+CAPTCHA STRATEGY (escalation order):
+1. First try headless — many sites use invisible reCAPTCHA v3 that passes without a challenge.
+2. If a challenge appears, use recaptcha_solve + recaptcha_answer to solve it visually.
+3. If challenges keep looping (3+ attempts), the server IP is flagged. Tell the user to run "iframer proxy" in a separate terminal to route traffic through their residential IP. This is the ONLY CLI command you may suggest. Check proxy status with the "status" tool.
+
 MINIMIZE TOKEN USAGE: Don't screenshot every action. Only look when you need to decide what to do next.
-NEVER use bash/CLI. NEVER suggest CLI commands to the user. ALL interaction goes through these MCP tools.`,
+NEVER use bash/CLI (except "iframer proxy" when CAPTCHA fails). ALL interaction goes through these MCP tools.`,
 });
 
 // ─── Tool 0: status (call this first) ────────────────────────────────
@@ -122,6 +127,12 @@ server.tool(
         if (credData.ok) status.credentials = credData.domains;
       } catch {}
 
+      // Check proxy tunnel
+      try {
+        const proxyData = await apiGet("/proxy/status");
+        if (proxyData.ok) status.proxyTunnel = proxyData.active;
+      } catch {}
+
       return { content: [{ type: "text", text: JSON.stringify(status, null, 2) }] };
     } catch (err) {
       return errorResponse(`Error: ${err.message}`);
@@ -214,7 +225,8 @@ server.tool(
   "interactive_act",
   `Send an action to the interactive browser and get a screenshot URL of the result.
 
-Actions: click, human-click, human-type, fill, navigate, scroll, wait, evaluate, wait-for-selector, keyboard.
+Actions: click, human-click, human-type, fill, navigate, scroll, wait, evaluate, wait-for-selector, keyboard, type-code.
+type-code: Type a verification code into split digit inputs in one call. Pass the full code as value (e.g. "735948"). Optionally pass selector for the first input (defaults to input[type="tel"]).
 reCAPTCHA: recaptcha-click (click checkbox), recaptcha-select (select tiles by index), recaptcha-verify (click verify), recaptcha-info (get challenge metadata).
 
 Screenshots are saved as files and returned as URLs. Use Read tool to view them. For reCAPTCHA, individual tile image URLs are returned.`,
@@ -222,7 +234,7 @@ Screenshots are saved as files and returned as URLs. Use Read tool to view them.
     action: z.enum([
       "click", "human-click", "human-type", "fill",
       "navigate", "scroll", "wait", "evaluate",
-      "wait-for-selector", "keyboard",
+      "wait-for-selector", "keyboard", "type-code",
       "recaptcha-click", "recaptcha-select", "recaptcha-verify", "recaptcha-info",
     ]).describe("Action type to perform"),
     selector: z.string().optional().describe("CSS selector (for click, fill, human-click, human-type, wait-for-selector)"),
@@ -514,6 +526,107 @@ ALWAYS call this tool when credentials are needed but not stored. NEVER tell the
   }
 );
 
+// ─── Tool 11: recaptcha_solve ───────────────────────────────────────
+
+server.tool(
+  "recaptcha_solve",
+  `Start solving a reCAPTCHA. Clicks the checkbox and if a challenge appears, returns ALL tile images inline in one response along with the prompt text. Much faster than individual recaptcha-click + screenshot + recaptcha-info calls.
+
+If the reCAPTCHA is solved immediately (no challenge), returns solved: true.
+If a challenge appears, returns the prompt and all tile images. Then use recaptcha_answer with the tile indices to complete it.`,
+  {},
+  async () => {
+    try {
+      const data = await apiPost("/interactive/act", {
+        action: { type: "recaptcha-solve" },
+        screenshot: false,
+      });
+      if (!data.ok) return errorResponse(`Error: ${data.error}`);
+
+      const result = data.result;
+      if (result.solved) {
+        return { content: [{ type: "text", text: "reCAPTCHA solved automatically (no challenge needed)." }] };
+      }
+
+      // Build content blocks: prompt text + all tile images
+      const content = [];
+      content.push({
+        type: "text",
+        text: `reCAPTCHA Challenge: "${result.prompt}"\nGrid: ${result.rows}x${result.cols} (${result.tiles.length} tiles)\nSelect the correct tiles by index (0-${result.tiles.length - 1}) using recaptcha_answer.`,
+      });
+
+      for (const tile of result.tiles) {
+        if (tile.image) {
+          content.push({
+            type: "text",
+            text: `Tile ${tile.index}:`,
+          });
+          content.push({
+            type: "image",
+            data: tile.image,
+            mimeType: "image/jpeg",
+          });
+        }
+      }
+
+      return { content };
+    } catch (err) {
+      return errorResponse(`Connection error: ${err.message}. Is the API running at ${BASE_URL}?`);
+    }
+  }
+);
+
+// ─── Tool 12: recaptcha_answer ──────────────────────────────────────
+
+server.tool(
+  "recaptcha_answer",
+  `Submit tile selections and verify the reCAPTCHA in one call. Selects the specified tiles and clicks verify.
+
+If solved, returns success. If a new challenge appears (common with reCAPTCHA), returns the new prompt and tile images inline — just call recaptcha_answer again with new selections.`,
+  {
+    tiles: z.array(z.number()).describe("Tile indices to select (e.g. [0, 3, 6])"),
+  },
+  async ({ tiles }) => {
+    try {
+      const data = await apiPost("/interactive/act", {
+        action: { type: "recaptcha-answer", tiles },
+        screenshot: false,
+      });
+      if (!data.ok) return errorResponse(`Error: ${data.error}`);
+
+      const result = data.result;
+      if (result.solved) {
+        return { content: [{ type: "text", text: "reCAPTCHA solved!" }] };
+      }
+
+      // New challenge — return tiles inline
+      const content = [];
+      content.push({
+        type: "text",
+        text: `Not solved yet — new challenge: "${result.prompt || "unknown"}"\nGrid: ${result.rows}x${result.cols} (${result.tiles.length} tiles)\nSelect the correct tiles by index using recaptcha_answer again.`,
+      });
+
+      for (const tile of result.tiles) {
+        if (tile.image) {
+          content.push({
+            type: "text",
+            text: `Tile ${tile.index}:`,
+          });
+          content.push({
+            type: "image",
+            data: tile.image,
+            mimeType: "image/jpeg",
+          });
+        }
+      }
+
+      return { content };
+    } catch (err) {
+      return errorResponse(`Connection error: ${err.message}. Is the API running at ${BASE_URL}?`);
+    }
+  }
+);
+
 // ─── Start ───────────────────────────────────────────────────────────
 
 const transport = new StdioServerTransport();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "iframer-cli",
-  "version": "1.0.2",
+  "version": "1.0.4",
   "description": "CLI and MCP server for iframer — a headful/headless browser for AI agents",
   "type": "module",
   "main": "index.js",
@@ -25,6 +25,7 @@
     "url": "https://github.com/EduardoFazolo/iframer-agentic-browser"
   },
   "dependencies": {
-    "@modelcontextprotocol/sdk": "^1.12.0"
+    "@modelcontextprotocol/sdk": "^1.12.0",
+    "ws": "^8.20.0"
   }
 }