generic-mcp

Generic MCP Client Integration Guide

OpenBoard connects to any client or agent framework supporting the Model Context Protocol (MCP) over stdio.

---

1. Installation

Install OpenBoard globally via npm:

npm install -g openboard

---

2. OpenBoard Startup

Start the local OpenBoard web dashboard and server at any time:

openboard start

For AI agent operations via MCP, your MCP client launches openboard mcp directly over stdio.

---

3. Universal Stdio Server Parameters & Configuration

For any generic MCP client, configure the following parameters:

• Command / Executable: openboard (or absolute path to openboard binary)
• Arguments: ["mcp"]
• Transport: stdio
• Protocol Version: 2024-11-05 (JSON-RPC 2.0)

Generic JSON Configuration Format

{
  "mcpServers": {
    "openboard": {
      "command": "openboard",
      "args": ["mcp"]
    }
  }
}

Equivalent Shell Invocation

openboard mcp

---

4. Verification

Verify that your generic MCP client can initialize the connection and query available tools:

1. Send an initialize JSON-RPC request.
2. Send a tools/list request.
3. Confirm that all 13 tools are returned (list_boards, create_board, get_board, rename_board, duplicate_board, favorite_board, delete_board, restore_board, get_canvas_state, get_canvas_screenshot, create_shapes, update_shapes, delete_shapes).

---

5. Example Workflows & Prompts

Standard Multi-Step Interaction Workflow

User:
List my OpenBoard boards.

Agent:
[Calls list_boards tool]
Found boards: "System Design", "Roadmap".

User:
Create a board called "Auth Pipeline" and draw OAuth2 authorization code flow.

Agent:
[Calls create_board with { name: "Auth Pipeline" }]
[Calls create_shapes with Client, Auth Server, Resource Server shapes and sequential arrows]
Created components and connected OAuth flow.

User:
Inspect the canvas and return a visual screenshot.

Agent:
[Calls get_canvas_state and get_canvas_screenshot]
Returned semantic layout and SVG vector screenshot.