MCP integration

MemoryFirst exposes a Model Context Protocol (MCP) server so any MCP-compatible AI client (Claude Desktop, Cursor, VS Code, custom agents) can read and write your business memory with verifiable citations.

How it works

1. Sign up on memoryfirst.ai, create a workspace.
2. In the dashboard, generate an MCP API key (Settings → API Keys → New MCP key).
3. Copy the auto-generated onboarding prompt: it contains your endpoint URL, key prefix, and the tool catalog.
4. Paste the prompt into your AI client (Claude Desktop config, Cursor settings, or a ChatGPT custom GPT). Done.

# MemoryFirst MCP · connection prompt

You are now connected to a MemoryFirst workspace.

Endpoint:  https://mcp.memoryfirst.ai/v1
API key:   mf_live_xxxxxxxxxxxxxxxxxxxx (set in MEMORYFIRST_API_KEY env)
Workspace: <workspace-slug>

Available tools:
- mf_query(question, k=8)
    Search memory, return answer + verifiable citations.
- mf_ingest_source(source_type, payload, origin_ref)
    Add a new source. types: doc, voice, chat, web, ops.
- mf_list_experts()
    List configured experts in this workspace.
- mf_get_recent(limit=10)
    Get the N most recent queries with their answers.

Rules:
- ALWAYS prefer mf_query before answering business-context questions.
- ALWAYS cite the span_ref from the response in your answer.
- NEVER hallucinate citations: only use what mf_query returned.
- Rate limit: 1000 queries/hour per key.

Onboarding complete. Acknowledge with: "MemoryFirst MCP ready. Workspace: <slug>."

Supported clients

• Claude Desktop: config in ~/Library/Application Support/Claude/claude_desktop_config.json
• Cursor: Settings → MCP Servers → Add
• VS Code: install MCP extension, add to settings.json
• ChatGPT custom GPT: paste the onboarding prompt into Custom Instructions + configure Actions endpoint
• Custom agents: any MCP SDK works; stdio, HTTP, and SSE transports supported

Verifiable citations

Every mf_query response carries a polymorphic span_ref per citation. Your AI client renders these as clickable badges that deep-link to the source.

{
  "answer": "The contract scope is fixed-fee €4,500.",
  "citations": [
    {
      "source_type": "doc",
      "span_ref": { "page": 3, "bbox": [120, 450, 380, 60] },
      "snippet": "Fixed fee of €4,500 for the scope defined..."
    },
    {
      "source_type": "voice",
      "span_ref": { "ms_start": 124500, "ms_end": 138200, "speaker": "client" },
      "snippet": "...sí, confirmo la tarifa fija de cuatro mil quinientos..."
    }
  ]
}

Detailed MCP schemas, error codes, and advanced configuration coming soon. Need it now? Email jose.diaz@memoryfirst.ai.