Perk MCP
Perk MCP is Perk's Model Context Protocol (MCP) server for connecting AI clients to your Perk travel, spend, invoice, and event data. Learn what it does, when to use it instead of the Travel and Spend REST APIs, and the prerequisites to get started.
The Perk Model Context Protocol (MCP) server lets you connect an AI client to Perk and query your travel, spend, invoice, and event data through MCP tool calls — putting answers inside the AI client you already use instead of requiring a manual report or a trip into the Perk app. This article explains what the Perk MCP server is, when to use it instead of the Perk REST APIs, and the key concepts you need before you connect a client.
TipNot a developer? Check out more about the Perk MCP and how to connect it to your AI assistant in the Perk Help Center.
What the Perk MCP server is
The Perk MCP server exposes Perk's travel and spend tools to Model Context Protocol (MCP)-compatible AI clients. MCP is an open standard that lets AI clients call tools and read data from external systems over a standard protocol.
The server connects an AI client — for example Claude Code, Cursor, Codex, or any MCP client — to Perk. Once connected, the client can call tools that read trips, expenses, travel invoices, reporting, policy, events, cards, and user data. You can ask questions in natural language and the client translates them into MCP tool calls against Perk.
You can query your data and generate reports. Write actions — booking, modifying, or canceling trips, creating, editing, or sending expenses, and approvals — are coming in a future release.
When to use the Perk MCP server
Use the Perk MCP server when you want an AI client or agent to query Perk in natural language or through MCP tool calls. It suits interactive, conversational, and agent-driven workflows where a person or an assistant explores Perk data on demand.
Use the Perk Travel API or Spend API (both REST) when you build a traditional coded integration — for example an enterprise resource planning (ERP) sync or a scheduled data pipeline. The REST APIs suit high-volume and scheduled workloads.
The table below summarizes when to choose each option.
| Choose | When you want to |
|---|---|
| Perk MCP server | Let an AI client or agent query Perk in natural language or through MCP tool calls |
| Travel API or Spend API (REST) | Build a coded integration such as an ERP sync or a scheduled, high-volume data pipeline |
Key concepts
The Perk MCP server builds on a few concepts you need before you connect a client. The following definitions make each concept self-contained.
- Tools are the individual actions the server exposes to an AI client. Each tool accesses a specific kind of Perk data — for example, a tool that lists trips or a tool that queries expenses. For the full list, see the Perk MCP tools reference.
- Scopes control which resources an access token can read. Perk uses colon-separated scope strings such as
user:readandtrip:read. A client requests the scopes it needs during sign-in, and you approve them in the browser. - Read and reporting scope describes what this release covers. You can query trips, expenses, invoices, reporting, policy, events, and users, and you can generate reports. Reporting data is one day behind because it goes through quality checks and aggregation; expense queries return live data.
Prerequisites
You need three things before you connect a client to the Perk MCP server.
- A Perk account with your usual Perk sign-in credentials.
- An MCP-compatible client, such as Claude Code, Cursor, Codex, or the MCP Inspector.
- Access to the data you want to query. Your access is governed by role-based access control (RBAC) — the access token is scoped to your role and your company's active plan, so a client can only read what you can already access in Perk.