Overview

Ramp MCP Server is an implementation of a Model Context Protocol (MCP) server designed to bring LLM-powered spend analysis to developer workflows. It accepts structured spend data (transactions, accounts, vendors) and exposes an API that uses language models and a set of programmatic tools to produce both human-readable insights and machine-actionable recommendations for cost optimization.

For engineering teams and finance platforms, Ramp MCP Server provides a simple integration point to get automated expense summaries, uncover savings opportunities, and generate prioritized recommendations. It’s useful when you want LLM reasoning over financial data while keeping tools and data access explicit and auditable through the MCP toolset.

Features

LLM-powered natural-language analysis of spend data
Structured, actionable recommendations (savings estimates, vendor consolidation, policy gaps)
MCP tool integration: charting, time-series aggregation, vendor grouping, anomaly detection
Extensible toolset so teams can add custom connectors or heuristics
Support for batched transaction ingestion and incremental analyses
Developer-friendly API and configuration (local or containerized deployment)
Human-readable explanations and references to the underlying data that led to suggestions

Installation / Configuration

Prerequisites: Node.js (16+), Docker (optional), API key(s) for your LLM provider if using hosted models.

Clone the repository:

git clone https://github.com/ramp-public/ramp-mcp.git
cd ramp-mcp

Install dependencies:

npm install
# or
yarn install

Create a .env file (example):

# .env
PORT=8080
RAMP_API_KEY=replace_with_your_ramp_api_key
LLM_API_KEY=replace_with_your_llm_provider_key
LLM_PROVIDER=openai  # or your provider name
LOG_LEVEL=info

Start the server locally:

npm run dev
# or for production
npm start

Run with Docker:

docker build -t ramp-mcp .
docker run -e RAMP_API_KEY=$RAMP_API_KEY -e LLM_API_KEY=$LLM_API_KEY -p 8080:8080 ramp-mcp

Modify environment variables to match your deployment and secret management practices.

Available Resources

Repository: https://github.com/ramp-public/ramp-mcp — source, examples, and tests
Configuration: .env-driven configuration for local, CI, and containerized deployments
Extensibility: add or replace MCP tools by implementing the tool interface in the tools/ directory (see repo docs)
API docs: example OpenAPI or API specification lives in the repo (look for api/ or docs/ folders)
Example datasets: sample transaction CSV/JSON under examples/ for local testing

Typical API Usage (example)

Below is a generic example of how to call the Developer API exposed by the server to request a spend analysis. Replace {MCP_BASE_URL} and keys with your values.

Request:

curl -X POST "{MCP_BASE_URL}/analyze" \
  -H "Authorization: Bearer $RAMP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "account_id": "acct_123",
    "period": { "from": "2024-01-01", "to": "2024-03-31" },
    "transactions": [
      { "date": "2024-02-05", "amount": 420.00, "vendor": "CloudCo", "category": "infrastructure" },
      { "date": "2024-02-07", "amount": 120.00, "vendor": "SaaSify", "category": "software" }
    ],
    "request": "summarize spend, highlight top optimization opportunities, and estimate monthly savings"
  }'

Example response (abridged):

{
  "summary": "Total spend $5,230 over 3 months. Top categories: infrastructure (42%), software (28%).",
  "insights": [
    { "id": "ins_1", "text": "CloudCo instances show 40% idle hours—consider resizing or reserved instances.", "estimated_monthly_savings": 600 },
    { "id": "ins_2", "text": "Duplicative SaaS subscriptions between teams A and B; cancel redundant seat licenses.", "estimated_monthly_savings": 320 }
  ],
  "actions": [
    { "type": "report", "url": "/reports/insights/2024-q1" }
  ]
}

Use Cases

Monthly CFO summary: ingest last month’s transactions and generate an executive summary with bulletized savings and outliers.
Vendor consolidation: identify multiple subscriptions to the same vendor across teams and recommend consolidation or enterprise licensing.
Anomaly detection and alerting: detect sudden spikes in category spend and automatically create an incident with contextual reasoning.
Policy enforcement: flag transactions that violate internal policies (e.g., high software spend without procurement approval) and suggest remediation steps.
Forecasting and scenario modeling: use historical data to estimate the impact of proposed cost reductions (e.g., switching to reserved instances).

Developer Tips

Keep input transactions normalized (ISO dates, consistent vendor naming) to improve LLM reasoning and grouping accuracy.
Use the tool extension points to add deterministic checks (e.g., contract lookup) so the LLM can call them instead of guessing.
Log requests and LLM decisions for auditability—responses often include recommended

Ramp MCP Server: LLM-Powered Spend Analysis