---
title: Ramp MCP
priority: 2
---
import { RyuNotice } from '@ramp/ryu'
import { MDXCodeBlocks } from '~/src/components/MDXCodeBlocks'
<RyuNotice title="Ramp MCP Remote is in Beta" color='warning' iconType='bell'>
Ramp MCP Remote is subject to change. Feedback and questions can be sent to developer-support@ramp.com.
</RyuNotice>
## Overview
Ramp MCP Remote is a secure, cloud-hosted service that enables AI assistants (like Claude or ChatGPT) to analyze your Ramp financial data using natural language queries. Using the [Model Context Protocol (MCP)](https://modelcontextprotocol.io/introduction), it provides a seamless bridge between your Ramp data and AI tools.
**What you can ask:**
- "Show me our top 10 vendors by spend last quarter"
- "Which departments exceeded their budgets this month?"
- "Analyze our travel expenses by employee and location"
- "Find all pending bills over $10,000"
> **Looking for a local MCP version?** Check out [Ramp MCP Local](https://github.com/ramp-public/ramp_mcp) on GitHub for an open-source local MCP server.
<RyuNotice type="warning">
**Connection Note:** Remote MCP connections are early technology and may require multiple attempts. If authentication fails, restart your client and try again.
</RyuNotice>
---
## Environment Selection
⚠️ **Important:** We recommend using only one environment at a time.
- **Production:** `https://ramp-mcp-remote.ramp.com/mcp` - Live Ramp data
- **Demo:** `https://demo-ramp-mcp-remote.ramp.com/mcp` - Sample data for testing
---
## Setup Instructions
### Prerequisites
1. **Ramp Account**: Active Ramp account with admin-level access
2. **Compatible Client**: Claude Desktop, Claude Code, Cursor, or other MCP-compatible client
### Claude Code
<MDXCodeBlocks title="Add MCP server to Claude config">
```bash
# Production (live data)
claude mcp add --transport sse ramp-server https://ramp-mcp-remote.ramp.com/mcp
# Demo (sample data)
claude mcp add --transport sse ramp-demo-server https://demo-ramp-mcp-remote.ramp.com/mcp
```
</MDXCodeBlocks>
**Authentication:**
1. Run `/mcp` in your Claude Code session to start authentication flow
2. Complete OAuth login when browser opens
3. Start asking questions about your Ramp data
### Claude Desktop
**Configuration file location:**
- **macOS:** `~/Library/Application Support/Claude/claude_desktop_config.json`
- **Windows:** `%APPDATA%\Claude\claude_desktop_config.json`
**Setup steps:**
1. **Install Node.js** (if not already installed):
- **macOS:** `brew install node`
- **Windows:** Download from [nodejs.org](https://nodejs.org)
2. **Open your configuration file:**
- **macOS:** Open Finder → Press 'Go' → 'Go to folder' → Enter the macOS path above
- **Windows:** Use the Windows path above in File Explorer
3. **Add to your configuration file:**
<MDXCodeBlocks title="Claude Desktop Configuration">
```json
{
"mcpServers": {
"ramp-mcp": {
"command": "npx",
"args": ["mcp-remote", "https://ramp-mcp-remote.ramp.com/mcp", "3334"]
}
}
}
```
</MDXCodeBlocks>
**For the demo environment, use:**
```json
"args": ["mcp-remote", "https://demo-ramp-mcp-remote.ramp.com/mcp", "3334"]
```
4. **Restart Claude Desktop** - Authentication will start automatically
### Cursor
**Configuration file location:**
- `~/.cursor/mcp.json`
**Setup steps:**
1. **Open your configuration file:**
- **macOS:** Open Finder → Press 'Go' → 'Go to folder' → Enter `~/.cursor/mcp.json`
- **Windows:** Navigate to `%USERPROFILE%\.cursor\mcp.json` in File Explorer
- **Create the file if it doesn't exist**
2. **Add to your configuration file:**
<MDXCodeBlocks title="Cursor Configuration">
```json
{
"mcpServers": {
"ramp-mcp-remote": {
"url": "https://ramp-mcp-remote.ramp.com/mcp"
}
}
}
```
</MDXCodeBlocks>
**For the demo environment:**
```json
{
"mcpServers": {
"demo-ramp-mcp-remote": {
"url": "https://demo-ramp-mcp-remote.ramp.com/mcp"
}
}
}
```
### Other MCP Clients
For clients supporting remote MCP connections, use these endpoints:
- **Production:** `https://ramp-mcp-remote.ramp.com/mcp`
- **Demo:** `https://demo-ramp-mcp-remote.ramp.com/mcp`
Refer to your client's MCP documentation for specific configuration steps.
---
## Getting Started
### Authentication & Setup
1. **Configure your client** using the setup instructions above
2. **Restart your client** after making configuration changes
3. **Authenticate with Ramp** - a browser window opens automatically for OAuth login
4. **Authorize permissions** for the data types you want to analyze
5. **Start asking questions** about your Ramp data
**Note:** Authentication happens automatically - you don't need to create OAuth applications or manage credentials.
---
## Troubleshooting
### Connection Issues
**Authentication fails or connection drops:**
- Restart your client and try connecting again
- Clear MCP authentication: `rm -rf ~/.mcp-auth` (if applicable)
- Ensure you're logged into the correct Ramp account
**"ETL operation limit reached":**
- Wait for current data operations to complete
- Ask the AI to clear unused tables to free up resources
**"Query returned more than 100 rows":**
- The AI will automatically add more specific filters and retry
### Common Setup Issues
**Configuration not working:**
- Verify JSON syntax in configuration files
- Ensure file paths are correct for your operating system
- Restart your client after making configuration changes
**Node.js or npm issues:**
- Ensure Node.js is properly installed
- Try `npm install -g mcp-remote` to install the required package
### Getting Help
- **Technical Support:** developer-support@ramp.com
- **API Documentation:** [Ramp Developer Docs](https://docs.ramp.com/developer-api)
- **MCP Protocol:** [Model Context Protocol](https://modelcontextprotocol.io/introduction)
---
## Available Tools
### Database Tools
Tools the AI uses to query and manage data after it's loaded:
1. **`execute_query`** - The AI runs SQL queries on loaded data
2. **`clear_table`** - The AI clears specific tables from memory
### Data Loading Tools
Tools the AI can use to load different types of Ramp data for analysis. All data loading happens asynchronously - the AI receives confirmation that loading has started and the tool will give status updates until data is ready to query.
| Tool | Description | Required Scope |
| -------------------------- | ---------------------------------------------------------------------------------------- | -------------------------------------------------------- |
| **`load_spend_export`** | **Recommended**: Load comprehensive spend data (transactions, reimbursements, and bills) | `transactions:read`, `reimbursements:read`, `bills:read` |
| **`load_purchase_orders`** | Load purchase order data | `purchase_orders:read` |
| **`load_cards`** | Load card information | `cards:read` |
| **`load_limits`** | Load spending limit data | `limits:read` |
| **`load_entities`** | Load entity information | `entities:read` |
| **`load_departments`** | Load department data | `departments:read` |
| **`load_users`** | Load user/employee data | `users:read` |
| **`load_memos`** | Load memo data | `memos:read` |
| **`load_locations`** | Load office location data | `locations:read` |
| **`load_spend_programs`** | Load spend program data | `spend_programs:read` |
| **`load_vendors`** | Load vendor information | `vendors:read` |
**Note:** Loading data into an existing table will replace all previously loaded data in that table
### Utility Tools
1. **`get_ramp_categories`** - Get all available Ramp spending categories (includes 40+ categories like Airlines, Restaurants, Office Supplies, etc.)
2. **`get_current_user`** - Get the current user's information (email, name, etc.)
3. **`submit_feedback`** - Allow the AI to submit feedback to Ramp about the MCP server interface or problems it runs into
---
## Usage Examples
### Analyzing Recent Spending
```
You: "What were our top expenses last month?"
Assistant: I'll analyze your expenses from last month. Let me load the necessary data.
[Behind the scenes: The AI loads transaction data from the previous month]
Loading data into table 'transactions' asynchronously...
Once the data is available, I'll run a query to identify your top expenses:
[The AI executes: SELECT merchant_name, SUM(amount) as total_spend FROM transactions
GROUP BY merchant_name ORDER BY total_spend DESC LIMIT 10]
Here are your top 10 expenses last month:
1. AWS Services - $45,230
2. Google Workspace - $12,500
3. Salesforce - $11,200
[... continued results]
```
### Department Budget Analysis
```
You: "Are any departments overspending this quarter?"
Assistant: I'll check department spending against budgets for this quarter.
[Behind the scenes: The AI loads multiple data sources]
- Loading transaction data for Q4 2024
- Loading department information
- Loading spending limits
Now I'll analyze spending by department:
[The AI monitors loading progress and runs queries once data is ready]
Based on my analysis:
- Marketing has spent $125,000 (104% of quarterly budget)
- Engineering is at $210,000 (87% of budget)
- Sales is at $95,000 (118% of budget - over by $14,000)
```
### Employee Spending Patterns
```
You: "I'm concerned about unusual employee expenses. Can you identify any outliers?"
Assistant: I'll analyze employee spending patterns to identify any unusual activity.
[Behind the scenes: The AI determines what data is needed]
- Loading user data
- Loading transaction data from the past 90 days
- Loading reimbursements
I'll now analyze spending patterns to find outliers:
[The AI runs statistical queries to identify unusual patterns]
I've identified some unusual patterns:
1. Employee John Smith had 15 restaurant charges totaling $3,200 last week (typical weekly average is $200)
2. Three employees submitted reimbursements over $5,000 (90th percentile is usually $1,500)
3. Unusual merchant category: "Casino" appeared 3 times for Employee Jane Doe
```
