Skip to content

Model Context Protocol (MCP) Tools Reference

Neural Memory provides an AI-native interface compliant with the Model Context Protocol (MCP) specification. This enables external AI agents (such as Claude Desktop, Cursor, Google Antigravity, and custom agent workflows) to autonomously query and reinforce your cognitive second brain.

1. Tool Categories & Schemas

1.1 Life & Personal Cognitive Tools

recall_decisions

Retrieves past strategic decisions, approvals, and rejections made by you or your team. - Arguments: - topic (optional, string): Filter decisions by specific topic keyword (e.g. "Cloud Q4"). - person (optional, string): Filter decisions involving a specific person (e.g. "Marco Rossi"). - project_id (optional, string, default: "default"): Project namespace. - limit (optional, integer, default: 10): Maximum results to return. - Example Response: 

[
  {
    "id": "d-a812f0c1",
    "title": "Approve Cloud Q4 Quote for 5000 EUR",
    "verdict": "APPROVED",
    "rationale": "User typed 'Ok procedi pure' in response to quote email",
    "timestamp": "2026-09-03T14:10:00Z",
    "person": "Marco Rossi"
  }
]

recall_commitments

Retrieves action items, deliverables, and promises with explicit deadlines and responsible parties. - Arguments: - status (optional, string, default: "open"): Filter by status ("open", "in_progress", "completed"). - debtor (optional, string): Who owes the deliverable ("user" or colleague name). - creditor (optional, string): Who is expecting delivery. - limit (optional, integer, default: 10): Maximum items. - Example Response: 

[
  {
    "id": "c-4b9981e2",
    "title": "Send updated slides by Friday 18:00",
    "due_date": "2026-09-05T18:00:00Z",
    "status": "open",
    "debtor": "user",
    "creditor": "Giulia Bianchi"
  }
]

recall_meetings

Retrieves summaries of collaborative video calls, meetings, and participant discussions. - Arguments: - topic (optional, string): Keyword search in meeting agendas or summaries. - attendee (optional, string): Colleague name who participated. - limit (optional, integer, default: 5). - Example Response: 

[
  {
    "title": "Architecture Sync with Roberto & Francesca",
    "start_time": "2026-09-03T10:00:00Z",
    "attendees": ["Francesca", "Roberto"],
    "summary": "Agreed to adopt hybrid SQLite standalone alongside Neo4j."
  }
]

get_daily_briefing

Generates an executive briefing summarizing all decisions, open commitments, and meetings for a target date. - Arguments: - date (optional, string, ISO format: YYYY-MM-DD): Target day (defaults to today). - project_id (optional, string, default: "default").

1.2 Project & Code Memory Tools

Performs deep hybrid search (full-text and graph neighborhood traversal) over the entire knowledge base. - Arguments: - query (required, string): Semantic search prompt. - limit (optional, integer, default: 10). - project_id (optional, string, default: "default").

get_project_context

Returns the aggregated active working context, recent activity slices, and concepts for a given project. - Arguments: - project_id (required, string): Target project identifier.

2. Client Configuration Guides

2.1 Claude Desktop Integration

Add the following configuration to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "neural-memory": {
      "command": "/usr/local/bin/kg-mcp",
      "args": ["--transport", "stdio"],
      "env": {
        "NEO4J_URI": "bolt://127.0.0.1:8787",
        "NEO4J_PASSWORD": "your-neo4j-password-from-env"
      }
    }
  }
}

Or run the automated configuration helper: 

./scripts/setup-mcp.sh --claude

2.2 Cursor & Antigravity IDE Integration

In Cursor settings under MCP Servers -> Add New MCP Server: - Name: neural-memory - Type: command - Command: kg-mcp --transport stdio

3. Real-World Agent Interaction Example

User to Claude Desktop:

"Claude, what did I promise to send to Giulia Bianchi this week, and did we agree on any cloud pricing today?"

Claude Autonomous Tool Execution: 1. Calls recall_commitments(creditor="Giulia Bianchi", status="open") $ o$ Returns: "Send updated slides by Friday 18:00" 2. Calls recall_decisions(topic="Cloud", limit=5) $ o$ Returns: "Approved Cloud Q4 Quote for 5000 EUR towards Marco Rossi"

Claude Response:

"You have an open commitment to send updated slides to Giulia Bianchi by Friday, September 5 at 18:00.

Regarding cloud pricing, earlier today you approved the Cloud Q4 Quote for 5,000 EUR communicated with Marco Rossi."