manus-ai-agent-integration（manus AI Agent 集成）

Manus AI Agent Integration

Use this skill when delegating complex autonomous tasks to Manus AI - an AI agent that can browse the web, execute code, generate files, and complete multi-step workflows independently.

When to Use Manus

• Research tasks: Market research, competitor analysis, fact-checking
• Content creation: Presentations, videos, websites, documents
• Data processing: PDF translation, data analysis, report generation
• Development: Chrome extensions, proof of concepts, prototypes
• Planning: Trip planning, fitness plans, business strategies

API Reference

Base URL

https://api.manus.ai/v1

Authentication

# Header authentication
API_KEY: your-api-key

Create Task

POST /v1/tasks
Content-Type: application/json
API_KEY: $MANUS_API_KEY

{
  "prompt": "Your task description",
  "agentProfile": "manus-1.5",  # manus-1.5 | manus-1.5-lite | speed | quality
  "taskMode": "agent",          # agent | chat | adaptive (optional)
  "projectId": "proj_xxx",      # optional
  "connectors": ["uuid1"],      # optional - gmail, calendar, notion
  "attachments": []             # optional - files, URLs, base64
}

Response:

{
  "task_id": "TeBim6FDQf9peS52xHtAyh",
  "task_title": "Generated Title",
  "task_url": "https://manus.im/app/TeBim6FDQf9peS52xHtAyh"
}

Agent Profiles

Create Project

POST /v1/projects

{
  "name": "Project Name",
  "instruction": "Default instruction for all tasks in this project"
}

Webhooks

Register webhook for real-time notifications:

POST /v1/webhooks
{"url": "https://your-server.com/webhook"}

Webhook Events:

• task_created - Task started
• task_progress - Task making progress (plan updates)
• task_stopped - Task completed or needs input

Webhook Payload (task_stopped):

{
  "event_type": "task_stopped",
  "task_detail": {
    "task_id": "xxx",
    "task_title": "...",
    "task_url": "https://manus.im/app/xxx",
    "message": "Completion message",
    "attachments": [
      {"file_name": "report.pdf", "url": "...", "size_bytes": 1234}
    ],
    "stop_reason": "finish"  # or "ask" if needs input
  }
}

CLI Usage

The manus CLI is installed at ~/.local/bin/manus:

# Set API key
export MANUS_API_KEY="sk-xxx"

# Create simple task
manus task "Summarize AI news from this week"

# Use quality mode for research
manus task -m quality "Research iOS 26 SwiftUI changes"

# Use playbook template
manus playbook use market-research "electric vehicles"
manus playbook use slide-generator "quarterly report"
manus playbook use video-generator "product demo"

# List all playbook templates
manus playbook list

# Create project with instruction
manus project create "Q4 Research" "Always cite sources"

# Task within project
manus task -p proj_xxx "Research task"

# Webhook management
manus webhook add https://my-server.com/hook
manus webhook remove webhook_id

Playbook Templates

Pre-built prompts for common use cases:

Integration Patterns

From Claude Code via CLI

# Execute via Bash tool
manus task -m quality "Research SwiftUI changes in iOS 26 and create a migration checklist"

Direct API Call

curl -X POST "https://api.manus.ai/v1/tasks" \
  -H "API_KEY: $MANUS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Analyze this codebase and suggest improvements",
    "agentProfile": "quality"
  }'

With Attachments

{
  "prompt": "Analyze this data",
  "attachments": [
    {"type": "url", "url": "https://example.com/data.csv"},
    {"type": "base64", "data": "...", "mimeType": "text/csv", "fileName": "data.csv"}
  ]
}

With Connectors

Available connectors (OAuth required at manus.im):

• Gmail - Email access
• Google Calendar - Calendar events
• Notion - Workspace access
• More at: https://manus.im/app?show_settings=integrations

{
  "prompt": "Schedule a meeting based on my calendar availability",
  "connectors": ["calendar-connector-uuid"]
}

Webhook Server Example

Simple Python webhook receiver:

from flask import Flask, request, jsonify

app = Flask(__name__)

@app.route('/webhook', methods=['POST'])
def handle_webhook():
    event = request.json
    event_type = event.get('event_type')

    if event_type == 'task_created':
        print(f"Task started: {event['task_detail']['task_id']}")

    elif event_type == 'task_progress':
        print(f"Progress: {event['progress_detail']['message']}")

    elif event_type == 'task_stopped':
        detail = event['task_detail']
        if detail['stop_reason'] == 'finish':
            print(f"Completed: {detail['message']}")
            for att in detail.get('attachments', []):
                print(f"  File: {att['file_name']} - {att['url']}")
        else:
            print(f"Needs input: {detail['message']}")

    return jsonify({"status": "ok"})

if __name__ == '__main__':
    app.run(port=8080)

Error Codes

Best Practices

1. Use quality mode for complex research tasks
2. Use projects to group related tasks with shared instructions
3. Set up webhooks for async task monitoring
4. Include context in prompts - Manus works better with specific details
5. Use playbooks for common workflows - they're optimized prompts

Hybrid Workflow (Claude Code + Manus)

Use the ai-delegate CLI to intelligently route tasks:

# Analyze where a task should go
ai-delegate analyze "scrape competitor reviews from App Store"
ai-delegate analyze "refactor BibleService to use async/await"

# Auto-route based on analysis
ai-delegate run "your task description"

# Force routing
ai-delegate manus "generate PDF report"
ai-delegate local "implement dark mode"

# See capability comparison
ai-delegate compare

Routing Logic

Use Claude Code (free, instant) for:

• Code generation, editing, refactoring
• File operations (read/write/search)
• Git operations, GitHub PRs
• Supabase database queries
• Web search, simple page fetching
• Apple docs via Sosumi MCP
• Memory persistence

Use Manus (costs credits) for:

• Browser automation (JS-rendered sites)
• Form filling, login-required scraping
• PDF/slide/video generation
• OAuth integrations (Gmail, Calendar)
• Multi-hour autonomous research
• Real purchases/bookings

Hybrid Example

Task: "Research competitor Bible apps and create a comparison report"

1. Claude Code: Web search for competitor list
2. Claude Code: Fetch public landing pages
3. Manus: Scrape App Store reviews (JS-rendered)
4. Claude Code: Analyze and structure data
5. Manus: Generate polished PDF report
6. Claude Code: Save to local files, commit to git

Resources

• API Docs: https://open.manus.ai/docs
• Playbook Gallery: https://manus.im/playbook
• API Settings: https://manus.im/app?show_settings=integrations&app_name=api
• Webhook Setup: https://manus.im/app?show_settings=integrations