MCP Protocol

Native integration for AI agents

What is MCP?

Model Context Protocol (MCP) is Anthropic’s standard for AI agents to use tools. Instead of HTTP requests, AI agents communicate directly via stdio with JSON-RPC 2.0.

Syncline supports both REST API (HTTP) and MCP Protocol (stdio). Functionality is identical—choose what fits your architecture.

Why Use MCP?

Zero Latency

No HTTP overhead. Direct process communication.

Native Integration

Works seamlessly with Claude and other MCP-compatible agents.

Stateful

Maintain context across multiple tool calls.

Simpler Auth

No API keys in headers. Auth via environment variables.

Architecture

┌─────────────────┐
│   AI Agent      │
│  (e.g., Claude) │
└────────┬────────┘
         │ stdio (JSON-RPC 2.0)
         │
┌────────▼────────┐
│ Syncline MCP    │
│     Server      │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  Google Calendar│
│    Firestore    │
└─────────────────┘

Available Tools

find_mutual_availability

Find time slots where attendees are available.

Input:

• attendees: Array of emails
• duration_minutes: Meeting duration (optional)

Output: 5 ranked time slots

schedule_meeting

Create a calendar event with Google Meet.

Input:

• attendees: Array of emails
• start_time: ISO 8601 datetime
• title: Meeting title
• duration_minutes: Duration (optional)

Output: Meeting details with Google Meet link

check_availability

Check a single user’s calendar.

Input:

• email: User’s email
• date_range: Start and end times

Output: Busy and free time slots

update_preferences

Update user scheduling preferences.

Input:

• email: User’s email
• preferences: New preferences object

Output: Updated preferences

Integration Guides

Claude Desktop

Integrate with Claude Desktop or Claude Code

Python Client

Build custom Python agents

Node.js Client

Build custom Node.js agents

Protocol Spec

Full JSON-RPC 2.0 specification

Quick Example

1. Start MCP Server

# Download Syncline binary
# Set environment variables
export SERVER_MODE=mcp
export GCP_PROJECT_ID=your-project
export GOOGLE_CLIENT_ID=...
export GOOGLE_CLIENT_SECRET=...
export API_ENCRYPTION_KEY=...

# Start in MCP mode
./syncline

2. Send JSON-RPC Request

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/list"
}

3. Receive Response

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "tools": [
      {
        "name": "find_mutual_availability",
        "description": "Find time slots where all attendees are available",
        "inputSchema": {
          "type": "object",
          "properties": {
            "attendees": {
              "type": "array",
              "items": {"type": "string", "format": "email"}
            }
          },
          "required": ["attendees"]
        }
      }
      // ... 3 more tools
    ]
  }
}

MCP vs REST API

Feature	MCP Protocol	REST API
Transport	stdio	HTTP
Format	JSON-RPC 2.0	REST JSON
Auth	Env vars	API key header
Latency	~1ms	~50ms
Streaming	Yes	No
Best For	AI agents	Web apps, mobile

Protocol Flow

1. Initialize

   Agent sends initialize method. Server responds with capabilities.

2. List Tools

   Agent calls tools/list. Server returns available tools with schemas.

3. Call Tool

   Agent calls tools/call with tool name and arguments. Server executes and returns result.

4. Repeat

   Agent can call tools multiple times in same session.

Example Conversation

Initialize

// Agent → Server
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "initialize",
  "params": {}
}

// Server → Agent
{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "protocolVersion": "2024-11-05",
    "serverInfo": {"name": "syncline", "version": "1.0.0"},
    "capabilities": {"tools": {"listChanged": false}}
  }
}

Next Steps

Integrate with Claude

Add Syncline to Claude Desktop

Build Python Client

Create custom Python agent