MCP Integration

Overview

MemLib includes a built-in Model Context Protocol (MCP) server that lets any MCP-compatible client — Claude Desktop, Cursor, Windsurf, and others — use your memories directly.

The MCP server uses Streamable HTTP transport at a single endpoint:

https://mem.anishroy.com/api/mcp

Authentication & Scope

The MCP server uses HTTP headers for authentication and scoping:

Locked vs Unlocked Mode

Locked mode is recommended for most use cases. It simplifies tool calls and prevents the AI from accidentally storing/querying the wrong scope.

Setup

1. Get Your API Key

From your project dashboard, go to API Keys and create a key with the appropriate permission level:

• Read — recall, prepare, diff, list only
• Read & Write — also allows storing memories
• Admin — full access

2. Configure Your Client

The MCP server expects auth and scope via headers, not query parameters. Here's how to configure each client:

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "memlib": {
      "url": "https://mem.anishroy.com/api/mcp",
      "headers": {
        "Authorization": "Bearer sk_your_api_key",
        "X-MemLib-Namespace": "my-app",
        "X-MemLib-Entity": "user-123"
      }
    }
  }
}

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "memlib": {
      "url": "https://mem.anishroy.com/api/mcp",
      "headers": {
        "Authorization": "Bearer sk_your_api_key",
        "X-MemLib-Namespace": "my-app",
        "X-MemLib-Entity": "user-123"
      }
    }
  }
}

Windsurf

Add to .windsurf/mcp.json:

{
  "mcpServers": {
    "memlib": {
      "url": "https://mem.anishroy.com/api/mcp",
      "headers": {
        "Authorization": "Bearer sk_your_api_key",
        "X-MemLib-Namespace": "my-app",
        "X-MemLib-Entity": "user-123"
      }
    }
  }
}

Unlocked Mode (no scope headers)

Omit the scope headers to let the AI specify namespace/entity on each tool call:

{
  "mcpServers": {
    "memlib": {
      "url": "https://mem.anishroy.com/api/mcp",
      "headers": {
        "Authorization": "Bearer sk_your_api_key"
      }
    }
  }
}

Tool Filtering

You can exclude specific tools using the excludeTools query parameter:

https://mem.anishroy.com/api/mcp?excludeTools=store_memory,diff_memories

This is useful if you want to limit what the AI can do — for example, exposing only read tools.

You can also configure this from the MCP page in your project dashboard by toggling tools on/off.

Available Tools

The MCP server exposes 5 tools:

store_memory

Store a memory with automatic fact extraction, deduplication, and conflict resolution.

Permission: Read & Write

Parameters:

• content (required) — the information to remember
• tags (optional) — tags for categorization
• source (optional) — origin identifier
• namespace / entity (only in unlocked mode)

recall_memories

Search memories by semantic similarity. Returns results ranked by hybrid score.

Permission: Read

Parameters:

• query (required) — natural language search query
• limit (optional) — max results
• category (optional) — filter by category
• min_importance (optional) — minimum importance score
• tags (optional) — filter by tags

list_memories

List stored memories.

Permission: Read

Parameters:

• limit (optional) — max results to return

prepare_context

Synthesize relevant memories into a context paragraph. Analyzes conversation intent, runs multi-query recall, and generates a briefing.

Permission: Read

Parameters:

• messages (required) — conversation messages to analyze
• max_candidates (optional) — max memories to consider

diff_memories

Memory changelog — what changed since a timestamp. Zero LLM calls.

Permission: Read

Parameters:

• since (required) — ISO timestamp
• limit (optional) — max events

Dashboard Config Builder

The easiest way to set up MCP is from the MCP page in your project dashboard. It provides:

• Client presets — select Cursor, Claude Desktop, Windsurf, or Generic
• Scope configuration — set namespace and entity
• Tool filtering — toggle tools on/off
• Test connection — verify your API key and endpoint work
• Copy config — one-click copy of the generated JSON config

Example Usage

Once configured, your AI assistant can use MemLib tools naturally:

User: "Remember that I prefer TypeScript and dark mode"

Claude: calls store_memory with content "I prefer TypeScript and dark mode"

"I've stored that! I'll remember your preference for TypeScript and dark mode."

User: "What do you know about my tech preferences?"

Claude: calls recall_memories with query "tech preferences"

"Based on what I remember: you prefer TypeScript, use dark mode, and your team uses React with Next.js."