MCP E-Sign for AI Agents: npx -y signbee-mcp Setup Guide

The Signbee MCP Tool Schema Specification

Unlike generic legacy document signing integrations that expose dozens of fragmented endpoints and require multi-step tab coordinate placements, Signbee MCP exposes two cleanly typed tools designed specifically for LLM tool calling accuracy.

JSON-RPC Stdio Protocol Trace

When Claude or Cursor triggers send_document, the client sends this JSON-RPC payload over stdio:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "send_document",
    "arguments": {
      "markdown": "# Non-Disclosure Agreement\n\n**Effective Date:** 2026-09-04\n\nBetween Acme Corp and Jane Doe...",
      "sender_name": "Acme Legal AI",
      "sender_email": "legal@acme.com",
      "recipient_name": "Jane Doe",
      "recipient_email": "jane@example.com",
      "expires_in_days": 7
    }
  }
}

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "Document successfully dispatched for signature!\nDocument ID: doc_7f3b891a2e\nSigning URL: https://signb.ee/sign/doc_7f3b891a2e\nAudit Hash: sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
      }
    ]
  }
}

Context Efficiency: Signbee MCP vs Alternative Implementations

In production agent architectures, every token injected into an agent's system prompt consumes valuable context window capacity and increases per-turn inference costs. When an agent exposes bloated tool suites, attention drift and tool hallucination rates rise sharply.

IDE & Client Configuration Walkthrough

{
  "mcpServers": {
    "signbee": {
      "command": "npx",
      "args": ["-y", "signbee-mcp"],
      "env": {
        "SIGNBEE_API_KEY": "your_api_key_here"
      }
    }
  }
}

Name: Signbee
Type: command
Command: npx -y signbee-mcp
Environment Variables: SIGNBEE_API_KEY=your_api_key_here

Remote Server-Sent Events (SSE) Transport for Distributed Swarms

While stdio is ideal for local desktop clients like Claude and Cursor, autonomous agent swarms running on container orchestrators (such as Kubernetes, AWS ECS, or Fly.io) require network-accessible endpoints. Signbee MCP natively supports Server-Sent Events (SSE) transport.

To host a shared Signbee MCP server across your internal engineering infrastructure:

# Run signbee-mcp as a daemonized SSE microservice on port 3001
docker run -d \
  --name signbee-mcp-cluster \
  -p 3001:3001 \
  -e SIGNBEE_API_KEY="sb_live_production_secret" \
  -e MCP_TRANSPORT="sse" \
  -e PORT="3001" \
  signbee/signbee-mcp:latest

Agents then configure their MCP client to connect over HTTP without spawning local child node processes:

{
  "mcpServers": {
    "signbee-remote": {
      "url": "https://mcp.internal.acme.com/sse",
      "transport": "sse",
      "headers": {
        "X-Shared-Secret": "cluster-internal-auth-token"
      }
    }
  }
}

Zero-Config Authentication: The Email OTP Fallback

Most developer tools require leaving your editor, generating API tokens in a web dashboard, and copying secrets into dotfiles. Signbee MCP removes this roadblock through automated Email OTP verification:

1. Agent Dispatches Without Key: If no SIGNBEE_API_KEY is detected, the MCP server contacts the Signbee API in unauthenticated developer mode.
2. Instant Verification Code: Signbee generates a 6-digit numeric OTP and delivers it to the sender_email address within 2 seconds.
3. Interactive Prompt: The agent pauses execution and prompts the user:
   “A verification code was sent to legal@acme.com. Please enter the 6-digit code to authorize this signature request.”
4. Authorization & Dispatch: Once entered, the contract is officially sealed and delivered to the recipient.

Production Hardening: Signals, Timeouts & Error Recovery

When integrating MCP servers into automated production pipelines, agents must gracefully recover from edge cases. The table below outlines common failure modes and recommended agent recovery policies:

Custom Python MCP Client Implementation

If you are building an autonomous multi-agent framework from scratch using Python, you can invoke signbee-mcp directly using the official Anthropic MCP SDK:

import asyncio
from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client

async def run_signbee_agent():
    # Define stdio server connection
    server_params = StdioServerParameters(
        command="npx",
        args=["-y", "signbee-mcp"],
        env={"SIGNBEE_API_KEY": "your_api_key_here"}
    )

    async with stdio_client(server_params) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()

            # Discover available tools
            tools = await session.list_tools()
            print("Discovered tools:", [t.name for t in tools.tools])

            # Execute send_document tool
            result = await session.call_tool(
                "send_document",
                arguments={
                    "markdown": "# Software Development Agreement\n\nScope of work...",
                    "sender_name": "Dev Studio",
                    "sender_email": "studio@dev.io",
                    "recipient_name": "Client Executive",
                    "recipient_email": "exec@client.io",
                    "expires_in_days": 10
                }
            )
            print("Tool execution response:", result.content[0].text)

if __name__ == "__main__":
    asyncio.run(run_signbee_agent())

Frequently Asked Questions