Vitally MCP Server

An MCP (Model Context Protocol) server that provides access to Vitally customer data via the Vitally API.

Features

• List customer accounts as resources

• Read account details

• Search for users by email or external ID

• Find accounts by name

• Query account health scores

• View account conversations and tasks

• Create notes for accounts

• Search through available tools

• Demo mode with mock data when no API key is provided

Setup for running locally

1. Install dependencies:

   npm install

2. Create a .env file in the root directory with the following:

   # Vitally API Configuration VITALLY_API_SUBDOMAIN=nylas # Your Vitally subdomain VITALLY_API_KEY=your_api_key_here # Your Vitally API key VITALLY_DATA_CENTER=US # or EU depending on your data center

3. Build the project:

   npm run build

Note: If you don't have a Vitally API key yet, the server will run in demo mode with mock data.

Getting your Vitally API Key

1. Navigate to your Vitally account

2. Go to Settings (⚙️) > Integrations > (new page) Vitally REST API

3. Toggle the switch to enable the integration

4. Copy the API Key (Secret Token)

Usage

There are three ways to use this MCP server:

Using the MCP Inspector

Run the MCP Inspector to test and debug the server:

This will open the MCP Inspector interface where you can interact with your server.

Running the MCP locally

1. First, find your Claude Desktop configuration file:

   • On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

   • On Windows: %APPDATA%\Claude\claude_desktop_config.json

2. Edit the config file to add the Vitally MCP server:

   { "mcpServers": { "vitally": { "command": "node", "args": ["--experimental-modules", "--experimental-specifier-resolution=node", "/Users/johnjung/nylas/vitally/vitally/build/index.js"] } } }

3. Restart Claude Desktop and you'll be able to use the Vitally MCP server.

Running the MCP via Docker

1. Edit the config file to add the Vitally MCP server from the GitHub package repository:

   { "mcpServers": { "vitally": { "command": "docker", "args": [ "run", "--rm", "-i", "-e", "VITALLY_API_SUBDOMAIN", "-e", "VITALLY_API_KEY", "-e", "VITALLY_DATA_CENTER", "ghcr.io/fiscaltec/vitally-mcp" ], "env": { "VITALLY_API_SUBDOMAIN": "VITALLY_API_SUBDOMAIN", "VITALLY_API_KEY": "VITALLY_API_KEY", "VITALLY_DATA_CENTER": "VITALLY_DATA_CENTER" } } } }

2. Restart Claude Desktop and you'll be able to use the Vitally MCP server.

Available Tools

Tool Discovery

• search_tools - Search for available tools by keyword

Account Management

• search_accounts - Search for accounts using multiple criteria (name, externalId)

• find_account_by_name - Find accounts by their name (partial matching supported)

• refresh_accounts - Refresh the cached list of accounts

• get_account_health - Get health scores for a specific account

User Management

• search_users - Search for users by email, external ID, or email subdomain

Communication & Tasks

• get_account_conversations - Get recent conversations for an account

• get_account_tasks - Get tasks for an account (can filter by status)

• create_account_note - Create a new note for an account

Example Questions to Ask

When connected to an MCP client like Claude, you can ask questions such as:

• "List all our customers"

• "Find accounts with 'Acme' in their name"

• "What's the health score for account X?"

• "Find user with email example@company.com"

• "Show me details about customer Y"

• "Get recent conversations for account Z"

• "What tasks are open for account A?"

• "Add a note to account B about our recent call"

• "What tools can I use for account management?"

Troubleshooting

• If you encounter JSON parsing errors, ensure you've removed all console.log statements from the code

• Make sure your .env file contains the correct API credentials

• Check that you've built the project (npm run build) after making changes

• Verify the path in claude_desktop_config.json is absolute and correct for your system

• If you don't have a valid API key, the server will run in demo mode with mock data

Attibution

As mentioned previously and in other files, this MCP has been created with original code from John Jung and containerised by Dan Searle.

All rightts go to John Jung.

Notes

• This has only been tested with Claude Desktop as of this moment however, is likely to woth with others but the configuration may not translate and is untested.

• Please raise an issue in either this repository, or the original if you find an issue or if you would like an improvement.