MCP Integration

Model Context Protocol integration for agentic coding environments

JSON-RPC 2.0

Node.js Bridge

Multi-Tenant

7 Tools

Overview

What is MCP Integration?

The MCP (Model Context Protocol) integration enables coding agents to access Cognitor directly. Through a lightweight Node.js bridge, agents can retrieve content, traverse pages, and run contextual search in realtime.

Architecture:

Agent Host (Cursor/Claude/other MCP clients) → Node.js Bridge → Cognitor API

Available Features

Content Operations

• Retrieve content elements and page trees
• Resolve structured context for agent tasks
• Multi-site and tenant-aware operations

Search & Planning

• Full-text search
• Search suggestions
• Site discovery for autonomous workflows

Setup & Installation

1. Download Bridge Script

Download:

cognitor_mcp_bridge.js📥 Download

2. Configure Your MCP Host

macOS Example (Claude Desktop):

File: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "cognitor-cms": {
      "command": "node",
      "args": ["/path/to/cognitor_mcp_bridge.js"]
    }
  }
}

Windows Example (Claude Desktop):

File: %APPDATA%\\Claude\\claude_desktop_config.json

{
  "mcpServers": {
    "cognitor-cms": {
      "command": "node",
      "args": ["C:\\path\\to\\cognitor_mcp_bridge.js"]
    }
  }
}

3. Restart MCP Host

After configuration, restart your MCP-compatible client so the bridge and tools are discovered correctly.

macOS:Cmd+Q → Reopen your client

Windows:Task Manager → End client process → Restart

Available Tools

discover_sites

Discovers available sites in the CMS

Parameters:

No parameters required

Example Question:

"What sites are available?"

content_list_elements

Lists content elements from a site

Parameters:

site_identifier (optional), limit, skip, type_filter

Example Question:

"Show me the latest articles"

content_get_element

Retrieves a single content element

Parameters:

element_id (required), site_identifier (optional)

Example Question:

"Get content element 123"

pages_list

Lists all pages from a site

Parameters:

site_identifier (optional), limit, skip

Example Question:

"Show me all pages"

pages_get

Retrieves a page with all content

Parameters:

page_id (required), site_identifier (optional)

Example Question:

"Load page 456 with all content"

search_content

Searches all content in a site

Parameters:

query (required), site_identifier, page, size

Example Question:

"Search for 'AI' in the content"

search_suggestions

Generates search suggestions

Parameters:

query (required), site_identifier, limit

Example Question:

"Give me search suggestions for 'tech'"

Practical Examples

How to Use MCP for Agentic Coding

Prompt examples for coding-agent workflows

🔍 Site Discovery

You: "What sites are available for this workspace?"

Agent: Uses discover_sites and scopes downstream tasks to the right site.

📄 Content Retrieval

You: "Show me the last 5 articles from the demo site"

Agent: Uses content_list_elements with automatic site detection.

🔎 Search

You: "Search all content for 'agentic coding'"

Agent: Uses search_content and returns ranked context with highlights.

📋 Page Management

You: "Load page 123 with all content elements"

Agent: Uses pages_get and returns a complete structured page graph.

Troubleshooting

Common Issues and Solutions

❌ "Server disconnected"

• Restart Claude Desktop

• Check Node.js installation: node --version

• Verify bridge script path in configuration

❌ "No tools available"

• Check configuration file syntax (valid JSON)

• Verify bridge script permissions

• Consult Claude Desktop logs

⚠️ "HTTP Error 404"

• Check backend availability

• Verify site identifier

• Use automatic site discovery

Additional Resources

API Documentation Live Demos MCP Protocol (GitHub)