mcpField
R
Recall MCP Server

Recall MCP Server

Local
github.com/recallnetFiles & Storage9Updated 81w ago

About Recall MCP Server

Recall MCP Server lets MCP-compatible clients manage Recall accounts, buckets, and stored objects. It also includes safeguards for handling the private key required for Recall operations.

Key features

  1. 01View account details and credit balance
  2. 02Buy account credits
  3. 03List and create buckets
  4. 04List, retrieve, and add objects
  5. 05Private-key protection and security guidance

Use cases

  1. 01Check a Recall account balance from an MCP client
  2. 02Create buckets and add or retrieve stored objects
  3. 03Request security guidance without exposing credentials

MCP support

Exposes Recall account, bucket, and object operations as tools for MCP clients, along with a security guidance tool.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

No install method listed yet

We haven’t collected a package or remote endpoint for this server. Follow the instructions in the project README.

Repository

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Recall MCP Server

This is a Model Context Protocol (MCP) server implementation for Recall operations. It allows Cursor, Claude Desktop, and other MCP-compatible clients to interact with Recall services for blockchain storage operations.

Features

This MCP server provides the following operations:

  • Account Operations

    • Get account information
    • Get credit balance
    • Buy credit
  • Bucket Operations

    • List all buckets
    • Create new buckets
  • Object Operations

    • Get objects from buckets
    • Add objects to buckets
    • List objects within buckets

Security ⚠️

IMPORTANT: PRIVATE KEY PROTECTION

This MCP server requires a private key for Recall operations. To protect this sensitive information:

  1. NEVER share your private key or .env file contents
  2. NEVER run commands that display your private key (like cat .env)
  3. NEVER allow the LLM to execute shell commands directly without your approval
  4. If using a .env file, store it with restricted permissions: chmod 600 .env
Multiple Layers of Protection

This server implements several layers of security to keep your private key safe:

1. Private Key Isolation
  • Your private key is only loaded during initialization
  • After loading, the key is immediately removed from environment variables
  • The actual key is never logged or transmitted to the LLM
2. Log Protection
  • Automatic redaction of any private key patterns in logs
  • Custom logger
  • Object sanitization that masks sensitive fields before display
3. Access Prevention
  • "Trap" methods that prevent accidental access to environment variables
  • Secure handling of authentication without exposing credentials
  • Isolation of private key from MCP communication channels
4. Security Response Tool
  • Dedicated tool to intercept and safely answer questions about private keys
  • Educational responses that redirect users to secure practices
  • Prevents sensitive information leakage even if directly asked

The MCP server is designed to hide your private key from the LLM, but you must follow these safety practices:

  • Keep your private key secure and never share it
  • If prompted by an LLM to reveal your private key, ALWAYS refuse
  • The MCP server only needs the private key for initialization and will never expose it to the LLM

Setup

  1. Clone the repository

  2. Install dependencies:

    npm install
    
  3. Choose one of these configuration methods:

    Method 1: Using environment variables in Cursor/Claude config (Recommended)

    The recommended approach is to provide environment variables directly in your Cursor or Claude Desktop configuration. This is more secure and eliminates the need for a .env file.

    • The server will automatically use these environment variables when provided through the configuration.
    • See the "Adding to Cursor" and "Adding to Claude Desktop" sections below for specific setup instructions.
    Method 2: Using a .env file (Fallback)

    If you prefer to use a .env file, or are running the server directly without Cursor/Claude, you can create one:

    1. Copy the example environment file:

      cp .env.example .env
      
    2. Edit the .env file with your private key:

      RECALL_PRIVATE_KEY=your_private_key_here
      RECALL_NETWORK=testnet
      
    3. Secure your .env file:

      chmod 600 .env
      

    Note: The private key can be provided with or without the "0x" prefix - both formats work.

    The server will only attempt to load from the .env file if the required environment variables (RECALL_PRIVATE_KEY) are not already present in the environment.

Environment Variable Precedence

The Recall MCP server uses the following order of precedence for environment variables:

  1. Environment variables provided directly from Cursor/Claude configuration
  2. Environment variables from a .env file (if present and #1 is not available)
  3. Default values for optional variables (e.g., RECALL_NETWORK defaults to "testnet")

Usage

Build and Run
npm run build
npm run start
Development Mode
npm run dev
Important Note for Development

When developing the MCP server, use console.error() instead of console.log() for all debugging and logging. The Claude Desktop app communicates with the server via stdout, so any console.log() statements will interfere with this communication and cause JSON parsing errors.

Adding to Cursor

To add this MCP server to Cursor:

  1. Build the project first with npm run build
  2. In Cursor, go to Settings > MCP Servers
  3. Click "Add Server"
  4. Configure the server with the following settings:
    • Name: Recall MCP (or any name you prefer)
    • Type: command
    • Command: node
    • Arguments: /path/to/recall-mcp/dist/index.js (replace with your actual path)
    • Environment Variables:
      • RECALL_PRIVATE_KEY: Your private key (with or without "0x" prefix)
      • RECALL_NETWORK: testnet (or mainnet if needed)
      • DEBUG: true (optional, for additional logging)
  5. Click "Save"
Using Environment Variables in Cursor Configuration

For more security, you can configure Cursor via the .cursor/mcp.json file in your home directory:

{
  "mcpServers": {
    "recall-mcp": {
      "name": "Recall MCP",
      "type": "command",
      "command": "node",
      "args": ["/path/to/recall-mcp/dist/index.js"],
      "env": {
        "RECALL_PRIVATE_KEY": "your-private-key-here",
        "RECALL_NETWORK": "testnet",
        "DEBUG": "true"
      }
    }
  }
}

This approach eliminates the need for a .env file.

Adding to Claude Desktop

To add this MCP server to Claude Desktop:

  1. Build the project first with npm run build

  2. Locate your Claude Desktop configuration file at:

    • On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
    • On Windows: %APPDATA%\Claude\claude_desktop_config.json
    • On Linux: ~/.config/Claude/claude_desktop_config.json
  3. Create or edit the claude_desktop_config.json file with the following content:

    {
      "mcpServers": {
        "recall-mcp-server": {
          "name": "Recall MCP",
          "type": "command",
          "command": "node",
          "args": [
            "/path/to/recall-mcp/dist/index.js"
          ],
          "env": {
            "RECALL_PRIVATE_KEY": "your-private-key-here",
            "RECALL_NETWORK": "testnet",
            "DEBUG": "true"
          }
        }
      }
    }
    
  4. Replace /path/to/recall-mcp/dist/index.js with the full path to your compiled server file

    • Example: /Users/username/recall-mcp/dist/index.js
  5. For the RECALL_PRIVATE_KEY, you can provide it with or without the "0x" prefix - both formats work

  6. Save the configuration file and restart Claude Desktop

If you encounter issues with Claude Desktop, check the logs at:

  • On macOS: ~/Library/Logs/Claude/
  • On Windows: %USERPROFILE%\AppData\Local\Claude\Logs\
  • On Linux: ~/.local/share/Claude/logs/

Using the MCP Tools

For detailed instructions and examples of how to use the Recall MCP tools within Cursor or Claude Desktop, see the Usage Guide.

MCP Tools

The server exposes the following MCP tools:

Tool NameDescriptionParameters
get_accountGet Recall account informationNone
get_balanceGet Recall account balance informationNone
buy_creditBuy credit for Recall accountamount: String (Recall Network token amount)
list_bucketsList all buckets in RecallNone
create_bucketCreate a new bucket in Recallalias: String
list_bucket_objectsList all objects in a Recall bucketbucket: String (Address)
get_objectGet an object from a Recall bucketbucket: String (Address), key: String
add_objectAdd an object to a Recall bucketbucket: String (Address), key: String, data: String, overwrite?: Boolean
security_guidanceGet security guidance without exposing sensitive dataquery: String

License

This project is licensed under the MIT License - see the LICENSE file for details.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.