🐘 Alephant MCP Server — AI FinOps Governance

Alephant MCP Server is a developer-centric infrastructure tool designed to integrate AI cost management directly into your AI workflow (such as Cursor, Claude Desktop, Windsurf, VS Code Copilot, and ChatGPT). By implementing the Model Context Protocol (MCP)—an open standard created by Anthropic—it transforms your AI assistant from a simple “code generator” into a “financially-aware architect”.

What is MCP and Why Alephant?

Think of MCP as a universal plugin system for AI. Just as USB-C allows you to connect any device to a computer with one cable, MCP allows any AI tool to connect to Alephant’s FinOps services through a standard protocol.

For Alephant, this means developers can interact with their Alephant dashboard using natural language directly within their IDE. There is no need to switch tabs and no need to log into a web backend.

Real-World Scenario:

A developer is coding in Cursor and wants to check their AI spending. They type: “How much did I spend on GPT-4 this week?”

Cursor calls the Alephant MCP server, which queries the Alephant API and responds: “You spent $47.23 on GPT-4 this week across 1,247 requests, a 12% decrease from last week.” — Instant answers directly within your workflow.

Core Capabilities & MCP Features

The Alephant MCP server exposes three main types of features to the AI Host:

Tools (Actions & Queries): Functions the AI can execute, such as fetching usage data, creating API keys, or setting alerts.
Resources (Read-only Data): Contextual data presented to the client, such as pricing lists or model catalogs.
Prompts (Templates): Pre-written templates, such as the “Cost Audit Report” prompt, which automatically pulls all relevant data.

🛠️ Key Tool Categories

Our MCP Server provides a comprehensive suite of tools categorized into four main areas:

Cost & Usage Tools (Highest Priority): Get usage summaries (get_usage_summary), daily costs (get_daily_costs), costs by model (get_cost_by_model), and retrieve request logs.
Key Management Tools: Manage virtual keys seamlessly. Includes listing keys (list_virtual_keys), creating keys with budgets (create_virtual_key), updating budgets (update_key_budget), and revoking keys (revoke_virtual_key).
Model & Routing Tools: Leverage Alephant’s smart routing. List available models (list_available_models), check model status, and update fallback configurations.
Alerts & Settings: Proactively set budget alerts when spending reaches specific thresholds.

Two Execution Modes

Based on the environment variables provided, the server operates in two distinct permission modes:

Mode	Environment Variables	Permissions
Virtual Key (VK)	`ALEPHANT_VIRTUAL_KEY`	Read-only, strictly scoped to a single cockpit
Manager	`ALEPHANT_PAT` + `ALEPHANT_WORKSPACE_ID`	Full workspace management and write access

Note: PAT takes precedence. If neither is set, the process will exit (Mock data is not used).

Transport Options

The Alephant MCP Server supports two standard transport layers:

Stdio (Local): Runs on your local machine and communicates via stdin/stdout. Ideal for local development, personal setups, and CLI tools. You can install it easily via npm/npx.
Streamable HTTP (Remote): Runs on a server with bidirectional communication via SSE/HTTP. Designed for team deployment, production environments, and hosted services (e.g., mcp.alephant.ai/mcp).

The Big Picture: The MCP server is not just a feature; it is a distribution channel. It brings the core value of Alephant—AI FinOps and governance—directly to where developers spend most of their time, eliminating friction and maximizing productivity.