204 MCP Tools in One Server: How Bridge ACE Gives AI Agents Real-World Superpowers
Most AI agent frameworks give you a way to call LLMs. Maybe they add file access. Maybe a web browser.
Bridge ACE ships with 204 MCP tools in a single 12,667-line Python file. Every agent gets access to all of them. Here's what that actually means.
The Tool Categories
Communication (20+ tools)
Agents can send and read messages across every major platform:
# Agent sends an email (requires human approval)
bridge_email_send(
to="client@company.com",
subject="Weekly Status Report",
body="..."
)
# Agent reads Slack messages
bridge_slack_read(channel="general", limit=10)
# Agent sends a WhatsApp message
bridge_whatsapp_send(to="+49...", message="Update: deployment complete")
# Agent posts to Telegram
bridge_telegram_send(chat_id="team_channel", text="Build passed ✓")
Phone & Voice (5 tools)
Yes, agents can make phone calls:
# Agent calls a number (Twilio + ElevenLabs voice)
bridge_phone_call(to="+49...", message="Your appointment is confirmed for tomorrow at 3pm")
# Agent listens to an active call
bridge_phone_listen(call_id="...")
# Agent speaks during a call
bridge_phone_speak(call_id="...", text="How can I help you today?")
Every phone call requires human approval via the Approval Gate system.
Browser Automation (40+ tools)
Three browser engines, from standard to stealth:
Playwright — Standard automation:
bridge_browser_navigate(url="https://example.com")
bridge_browser_click(selector="#submit")
bridge_browser_fill(selector="input[name='email']", value="...")
Stealth Browser — Passes bot detection (Cloudflare, DataDome):
bridge_stealth_start(stealth_level="high")
bridge_stealth_goto(url="https://protected-site.com")
bridge_stealth_fill(selector="input", value="...")
CDP — Attach to user's real Chrome:
bridge_cdp_connect(port=9222)
bridge_cdp_navigate(url="https://linkedin.com/messaging")
# Full access to user's logged-in sessions
Desktop Control (18 tools)
Full desktop automation — screenshot, click, type, scroll:
bridge_desktop_screenshot() # Captures entire screen
bridge_desktop_click(x=500, y=300) # Click at coordinates
bridge_desktop_type(text="Hello World") # Type text
bridge_desktop_key(combo="ctrl+s") # Keyboard shortcuts
bridge_desktop_window_focus(window_name="Chrome") # Window management
Agent Coordination (25+ tools)
The tools that make multi-agent teams work:
# Agent registers with the platform
bridge_register(agent_id="backend", role="Backend Developer")
# Agent sends a message to another agent
bridge_send(to="frontend", content="API endpoint /users is ready")
# Agent locks files it's working on
bridge_scope_lock(paths=["src/api/"], reason="Building user endpoints")
# Agent requests human approval
bridge_approval_request(action="Send deployment email to all@company.com")
# Agent creates a task for another agent
bridge_task_create(title="Write unit tests for /users endpoint", assigned_to="qa")
Knowledge & Memory (15 tools)
Persistent knowledge across sessions:
# Agent writes to its knowledge vault
bridge_knowledge_write(path="decisions/api-design.md", content="...")
# Agent searches memory
bridge_memory_search(query="how did we handle auth last time?")
# Agent stores credentials securely (Fernet encrypted)
bridge_credential_store(service="github", key="api_token", value="...")
Creator Pipeline (20 tools)
Video, audio, and content creation:
bridge_creator_url_ingest(url="https://youtube.com/watch?v=...")
bridge_creator_voiceover(text="Welcome to Bridge ACE", voice="...")
bridge_creator_export_clip(start=10, end=45, output="clip.mp4")
bridge_creator_publish(platform="youtube", title="...", video="clip.mp4")
The Approval Gate System
Every tool is classified by risk:
| Classification | Behavior | Examples |
|---|---|---|
| AUTO | Execute immediately | File read, code analysis, internal messaging |
| LOG | Execute and log | Web searches, research |
| REQUIRE_APPROVAL | Queue for human approval | Email send, phone call, purchases, git push |
When an agent tries to send an email, the UI shows an approval popup. You approve or deny. Only then does the action execute.
All In One Server
All 204 tools run from a single bridge_mcp.py file using the MCP stdio transport. Every Claude Code, Codex, Qwen, or Gemini CLI agent that connects gets access to all tools automatically.
No separate installations. No microservices. No Docker compose with 15 containers.
# Start the platform — all tools are available
./Backend/start_platform.sh
Open Source
Apache 2.0. Self-hosted. Every tool, every line of code.
Top comments (0)