@inetafrica/open-claudia 2.2.19 → 2.2.20

package/Dockerfile

@@ -1,6 +1,7 @@
 FROM node:20-slim
 
 # Install system dependencies
+# chromium is the browser engine driven by agent-browser (apt pulls its headless libs).
 RUN apt-get update && apt-get install -y --no-install-recommends \
     curl \
     ffmpeg \
@@ -13,6 +14,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     sudo \
     openssh-client \
     rsync \
+    chromium \
     && rm -rf /var/lib/apt/lists/*
 
 # Install Claude Code CLI
@@ -22,6 +24,13 @@ RUN curl -fsSL https://claude.ai/install.sh | sh || \
 # Install OpenAI Codex CLI for the /codex backend
 RUN npm install -g @openai/codex
 
+# Browser automation for the agent: vercel-labs agent-browser (Rust CLI, token-efficient,
+# replaces Playwright MCP). Installed at build time as root so the non-root runtime user
+# doesn't hit EACCES on `npm install -g`. Chrome is pinned to the apt chromium binary so it
+# lives outside the /data VOLUME (build-time writes to /data are discarded at runtime).
+RUN npm install -g agent-browser
+ENV AGENT_BROWSER_EXECUTABLE_PATH=/usr/bin/chromium
+
 # Create non-root user (Claude Code refuses --dangerously-skip-permissions as root)
 # node:20-slim already has uid/gid 1000 (node user). Create claudia with different IDs.
 RUN groupadd -g 1001 claudia && useradd -u 1001 -g 1001 -m -d /data claudia

package/core/system-prompt.js

@@ -182,6 +182,20 @@ Sub-agents (spawn a fresh throwaway Claude for focused research — output comes
 - Use when a side question would pollute this conversation, or to fan out independent lookups.
 - The sub-agent has Read/Glob/Grep/Bash but no access to your chat session or send-* tools.
 
+## Browser automation
+You can drive a real web browser via the \`agent-browser\` CLI (vercel-labs). It is a plain command you run through Bash — NOT an MCP tool, so there is nothing to enable; just call it. Chrome is bundled in the image and runs headless (it auto-adds --no-sandbox in this container). Use it to read pages, fill forms, click through flows, scrape content, or capture screenshots of how something renders.
+
+Core workflow — open, snapshot to get element refs, then act on those refs:
+- \`agent-browser open <url>\` — navigate (also: \`goto\`, \`back\`, \`forward\`, \`reload\`).
+- \`agent-browser snapshot -i\` — list interactive elements as @refs (e.g. @e1, @e2). Do this before clicking/filling so you target the right element. Add \`--json\` for machine-readable output.
+- \`agent-browser click @e1\` / \`fill @e2 "text"\` / \`type\`, \`press\`, \`select\`, \`hover\`, \`check\`/\`uncheck\`, \`upload\` — interact.
+- \`agent-browser get text @e3\` / \`get html\` / \`get url\` / \`get title\` / \`is visible @e4\` — read state.
+- \`agent-browser screenshot <path.png>\` — capture; then deliver with \`open-claudia send-photo <path.png>\`.
+- \`agent-browser eval "<js>"\` — run JS in the page when no command fits.
+- \`agent-browser close\` — end the session when done.
+
+Notes: a background daemon persists between commands, so navigation state carries across calls in a turn. Prefer \`snapshot -i\` refs over guessing selectors. Keep output lean by acting on refs rather than dumping full HTML. This replaced the old Playwright MCP — do not look for a playwright tool.
+
 If you tell the user "I'll check back in N minutes" or "I'll run this every morning", you MUST schedule it with one of the commands above in the same turn. Otherwise nothing happens.
 
 ## Guidelines

package/health.js

@@ -336,32 +336,15 @@ async function runHealthChecks(options = {}) {
     results.warnings.push(`FFmpeg not found at: ${env.FFMPEG} (voice notes disabled)`);
   }
 
-  // 12. Check Playwright MCP setup
-  try {
-    const root = execSync("npm root -g", { encoding: "utf-8", stdio: "pipe" }).trim();
-    const pwMcpPkg = path.join(root, "@playwright", "mcp", "package.json");
-    if (!fs.existsSync(pwMcpPkg)) {
-      results.warnings.push("@playwright/mcp not installed globally (browser tools unavailable)");
-    } else {
-      results.checks.playwright_mcp = { ok: true };
-    }
-  } catch (e) {
-    results.warnings.push("Could not check @playwright/mcp installation");
-  }
-
-  if (fs.existsSync(MCP_CONFIG_FILE)) {
-    try {
-      const mcpCfg = JSON.parse(fs.readFileSync(MCP_CONFIG_FILE, "utf-8"));
-      if (!mcpCfg.mcpServers || !mcpCfg.mcpServers.playwright) {
-        results.warnings.push("Playwright MCP not configured in ~/.claude/mcp.json");
-      } else {
-        results.checks.mcp_config = { ok: true };
-      }
-    } catch (e) {
-      results.warnings.push("~/.claude/mcp.json is malformed");
-    }
+  // 12. Check browser automation (agent-browser CLI + Chrome)
+  if (!binaryExists("agent-browser")) {
+    results.warnings.push("agent-browser CLI not found on PATH (browser tools unavailable)");
   } else {
-    results.warnings.push("~/.claude/mcp.json not found (MCP tools unavailable)");
+    results.checks.agent_browser = { ok: true };
+    const chrome = env.AGENT_BROWSER_EXECUTABLE_PATH;
+    if (chrome && !fs.existsSync(chrome)) {
+      results.warnings.push(`Chrome not found at AGENT_BROWSER_EXECUTABLE_PATH=${chrome} (browser tools may fail)`);
+    }
   }
 
   return results;
@@ -406,99 +389,27 @@ function formatResults(results, verbose = false) {
 
 // ── Idempotent setup: ensures required tools and config are in place ──
 
-const MCP_CONFIG_FILE = path.join(process.env.HOME || require("os").homedir(), ".claude", "mcp.json");
-
-/**
- * Ensure @playwright/mcp is installed globally.
- * Returns { installed: bool, alreadyInstalled: bool, error?: string }
- */
-function ensurePlaywrightMcp() {
-  // Check if already installed
-  try {
-    const root = execSync("npm root -g", { encoding: "utf-8", stdio: "pipe" }).trim();
-    const pkgPath = path.join(root, "@playwright", "mcp", "package.json");
-    if (fs.existsSync(pkgPath)) {
-      return { installed: true, alreadyInstalled: true };
-    }
-  } catch (e) { /* not installed */ }
-
-  // Install it
-  try {
-    execSync("npm install -g @playwright/mcp@latest 2>&1", {
-      encoding: "utf-8",
-      timeout: 120000,
-      stdio: "pipe",
-      env: { ...process.env, PATH: process.env.PATH },
-    });
-    return { installed: true, alreadyInstalled: false };
-  } catch (e) {
-    return { installed: false, error: (e.stderr || e.message || "").slice(-300) };
-  }
-}
-
 /**
- * Ensure ~/.claude/mcp.json has the Playwright MCP server configured.
- * Idempotent — won't duplicate if already present.
- * Returns { configured: bool, alreadyConfigured: bool, error?: string }
+ * Verify the agent-browser CLI is available for browser automation.
+ * The binary + Chrome are baked into the image at build time (see Dockerfile),
+ * so this is a presence check, not an installer — a non-root runtime user can't
+ * `npm install -g` anyway. Returns { installed, alreadyInstalled, error? }.
  */
-function ensureMcpConfig() {
-  const claudeDir = path.dirname(MCP_CONFIG_FILE);
-
-  // Ensure ~/.claude directory exists
-  if (!fs.existsSync(claudeDir)) {
-    try {
-      fs.mkdirSync(claudeDir, { recursive: true });
-    } catch (e) {
-      return { configured: false, error: `Failed to create ${claudeDir}: ${e.message}` };
-    }
-  }
-
-  // Load existing config or start fresh
-  let mcpConfig = { mcpServers: {} };
-  if (fs.existsSync(MCP_CONFIG_FILE)) {
-    try {
-      mcpConfig = JSON.parse(fs.readFileSync(MCP_CONFIG_FILE, "utf-8"));
-      if (!mcpConfig.mcpServers) mcpConfig.mcpServers = {};
-    } catch (e) {
-      // Corrupted config — back up and start fresh
-      const backup = MCP_CONFIG_FILE + ".bak." + Date.now();
-      try { fs.copyFileSync(MCP_CONFIG_FILE, backup); } catch (e2) {}
-      mcpConfig = { mcpServers: {} };
-    }
-  }
-
-  // Check if Playwright MCP is already configured
-  if (mcpConfig.mcpServers.playwright) {
-    return { configured: true, alreadyConfigured: true };
-  }
-
-  // Find the installed binary
-  let mcpBin = "playwright-mcp";
-  try {
-    const resolved = execSync("which playwright-mcp", { encoding: "utf-8", stdio: "pipe" }).trim();
-    if (resolved) mcpBin = resolved;
-  } catch (e) {
-    // Try npx path
-    try {
-      const root = execSync("npm root -g", { encoding: "utf-8", stdio: "pipe" }).trim();
-      const binPath = path.join(root, "@playwright", "mcp", "cli.js");
-      if (fs.existsSync(binPath)) mcpBin = binPath;
-    } catch (e2) {}
-  }
-
-  // Add Playwright MCP server config
-  mcpConfig.mcpServers.playwright = {
-    command: "npx",
-    args: ["@playwright/mcp@latest", "--headless"],
-  };
-
-  // Write config
-  try {
-    fs.writeFileSync(MCP_CONFIG_FILE, JSON.stringify(mcpConfig, null, 2) + "\n", "utf-8");
-    return { configured: true, alreadyConfigured: false };
-  } catch (e) {
-    return { configured: false, error: `Failed to write ${MCP_CONFIG_FILE}: ${e.message}` };
-  }
+function ensureAgentBrowser() {
+  if (!binaryExists("agent-browser")) {
+    return {
+      installed: false,
+      error: "agent-browser CLI not on PATH (expected from image build; browser tools unavailable)",
+    };
+  }
+  const chrome = process.env.AGENT_BROWSER_EXECUTABLE_PATH;
+  if (chrome && !fs.existsSync(chrome)) {
+    return {
+      installed: false,
+      error: `Chrome missing at AGENT_BROWSER_EXECUTABLE_PATH=${chrome}`,
+    };
+  }
+  return { installed: true, alreadyInstalled: true };
 }
 
 /**
@@ -509,15 +420,9 @@ function ensureMcpConfig() {
 function ensureSetup() {
   const results = { ok: true, steps: {} };
 
-  // Step 1: Install Playwright MCP globally
-  results.steps.playwright_mcp = ensurePlaywrightMcp();
-  if (!results.steps.playwright_mcp.installed) {
-    results.ok = false;
-  }
-
-  // Step 2: Configure MCP in Claude Code config
-  results.steps.mcp_config = ensureMcpConfig();
-  if (!results.steps.mcp_config.configured) {
+  // Browser automation: verify agent-browser CLI + Chrome are present.
+  results.steps.agent_browser = ensureAgentBrowser();
+  if (!results.steps.agent_browser.installed) {
     results.ok = false;
   }
 
@@ -554,8 +459,7 @@ module.exports = {
   binaryExists,
   ensureSetup,
   formatSetupResults,
-  ensurePlaywrightMcp,
-  ensureMcpConfig,
+  ensureAgentBrowser,
   REQUIRED_ENV_KEYS,
   OPTIONAL_ENV_KEYS,
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@inetafrica/open-claudia",
-  "version": "2.2.19",
+  "version": "2.2.20",
   "description": "Your always-on AI coding assistant — Claude Code, Cursor Agent, and OpenAI Codex via Telegram or Kazee Chat",
   "main": "bot.js",
   "bin": {