MCP Integration

The MCP (Model Context Protocol) server exposes Apart Intelligence as a set of tools that AI assistants can use directly. Your organizational knowledge becomes available inside Claude, Cursor, and any MCP-compatible tool — no copy-paste needed.

Setup

Quick example — add to your Claude Desktop or Cursor config:

claude_desktop_config.json / .cursor/mcp.json

{
  "mcpServers": {
    "apart-intelligence": {
      "command": "npx",
      "args": ["-y", "@apart-tech/intelligence-mcp"],
      "env": {
        "APART_API_URL": "https://api.intelligence.apart.tech",
        "APART_API_KEY": "apart_your-api-key"
      }
    }
  }
}

For Claude Code, install the CLI skills for the best experience:

ai codebase init

Tip

See the MCP Setup Guidefor detailed instructions for every supported tool — Claude Code, Cursor, VS Code, Windsurf, Codex CLI, Gemini CLI, Cline, JetBrains, and Zed — including tool-specific quirks, recommended approaches, and troubleshooting.

Available Tools

The MCP server provides 28 tools organized by category:

Read Tools

Tool	Description
`search_knowledge`	Hybrid semantic + keyword search with workspace filtering
`get_node`	Retrieve a specific node with its edges (supports PII decryption)
`get_context`	Assemble a context package via search + graph traversal
`get_neighborhood`	N-hop graph neighborhood of a node
`list_drafts`	List nodes pending review
`list_types`	Node and edge type inventory
`list_domains`	Full domain taxonomy tree
`list_workspaces`	All available workspaces
`get_workspace`	Workspace definition with filter rules
`export_workspace`	Export workspace as markdown

Write Tools

Tool	Description
`add_node`	Create a knowledge node (defaults to draft)
`update_node`	Modify an existing node
`link_nodes`	Create a typed edge between two nodes
`set_status`	Change lifecycle status (draft/approved/archived)
`delete_node`	Delete a node and all its edges
`delete_edge`	Delete an edge between two nodes

Graph Health Tools

Tool	Description
`health_report`	Comprehensive graph health report
`validate_edges`	Detect self-loops, duplicates, low-relevance connections
`find_orphans`	Find nodes with no connections
`find_duplicates`	Find near-duplicate nodes by embedding similarity
`find_islands`	Detect disconnected clusters with optional bridge suggestions
`suggest_links`	Suggest missing links between similar unconnected nodes
`analyze_relationship_types`	Find semantically similar relationship types to merge
`retype_edges`	Rename all edges of one type to another

PII Tools

Tool	Description
`get_pii_config`	Get the organization's PII detection/encryption policy
`test_pii`	Test text for PII detection

Maintenance Tools

Tool	Description
`seed_domains`	Seed the default domain taxonomy
`re_embed`	Re-generate embeddings for nodes

Using Workspaces with MCP

Both search_knowledge and get_context accept a workspaceparameter that automatically applies the workspace's domain and type filters:

// Inside Claude, the AI can call:
search_knowledge({
  query: "deployment process",
  workspace: "tasks-ops"
})

// Or get full context:
get_context({
  query: "assignment resolution",
  workspace: "tasks-engineering"
})

This ensures the AI gets role-appropriate context. An engineering workspace returns architecture docs and code. A product workspace returns feature descriptions.

Tip

When an AI tool has access to your knowledge graph via MCP, it can answer questions grounded in your actual architecture, processes, and decisions — instead of generating generic responses.

Configuration

The MCP server requires two environment variables to connect to your Apart Intelligence API:

Variable	Description
`APART_API_URL`	Your API server URL (e.g. `https://api.intelligence.apart.tech`)
`APART_API_KEY`	Your API key (starts with `apart_`)

These are passed via the env field in your MCP configuration, as shown in the setup examples above. Get your API key by running ai login with the CLI, or from the Apart Intelligence dashboard.