Expense MCP (beta)

Connect Findity expense functionality to Claude, ChatGPT, Gemini, or an agent you build yourself, using the Model Context Protocol.

This guide covers what the Findity Expense MCP server exposes, how to connect it from the major assistant clients, and what people realistically use it for. It is not an API reference — the MCP server is a surface over the Expense API, and the underlying endpoints are documented there.

What it is

The Model Context Protocol (MCP) is an open standard for connecting AI assistants to external systems. An MCP server publishes a set of callable tools; an MCP-compatible client — Claude, ChatGPT, Gemini, a coding agent, or something you build — discovers those tools and calls them on the user's behalf.

The Findity Expense MCP server publishes Findity's expense functionality as such a set of tools. A user asks their assistant a question in natural language, the assistant calls the relevant Findity tool, and the result lands in Findity exactly as if the user had done it in the app.

Two things follow from this that shape everything below:

• The server acts as one authenticated user. It does not introduce a new permission model. It reuses the same login, the same roles, and the same approval flows as any other Findity client.
• Partners can build against it, not just connect to it. An off-the-shelf assistant is the fastest route, but the same server backs a custom agent — which is what the Slack and Teams section is about.

This is a different server from the Docs MCP, which exposes Findity's API documentation to coding assistants. The Docs MCP helps a developer write an integration; the Expense MCP handles actual expenses.

Tools it exposes

This isn't a full API reference — see the Expense API docs for that — but it helps to know the shape of what's available before you start asking an assistant to use it.

What it can access

A session runs as the user who signed in, with exactly the access that user has in the Findity app.

Actions taken through the MCP route through the same roles and approval flows as any other client. Submitting an expense through an assistant puts it into normal approval routing; approving one through an assistant is the same approval the user could perform in the app.

Out of scope in beta

• Retrieving expenses across the organisation
• Acting on behalf of another user
• Forwarding approved expenses to external systems

Before you start

1. Confirm with Findity that the Expense MCP is enabled for your environment.
2. Have Findity credentials for the user who will be signing in. Access follows that user's existing role, so test with an account that has the permissions you want to demonstrate — including approver rights if you plan to try the approval flows.
3. Use an MCP-compatible client that supports remote servers over HTTP. Clients that only support locally-installed servers cannot reach a hosted endpoint.

Connect the server

Server URL

https://api.findity.com/api/mcp

White label partners use their own prefix in place of api:

https://{partner-prefix}.findity.com/api/mcp

There is nothing to install and no API key to issue. On first use the client opens a browser window and prompts for a Findity sign-in, exactly as the app does. Once signed in, the client holds the session and the tools become available to the assistant.

claude mcp add --transport http findity https://api.findity.com/api/mcp

{
  "mcpServers": {
    "findity": {
      "httpUrl": "https://api.findity.com/api/mcp",
      "authProviderType": "dynamic_discovery"
    }
  }
}

{
  "mcpServers": {
    "findity": {
      "url": "https://api.findity.com/api/mcp"
    }
  }
}

Verify the connection

Ask the assistant something that requires a tool call and has an unambiguous answer, such as:

What expenses are currently waiting for my approval?

A working connection produces an answer drawn from Findity, and the client shows the tool call it made. An assistant that answers in general terms without calling a tool is not connected — check that the connector is enabled for the current conversation, not just added to the account.

What you can ask it to do

The assistant decides which tools to call from what the user asks, so there are no commands to memorise. Two patterns are worth calling out separately, because they need different things from the person using them.

Everyday questions and actions

Single requests, answered in one or two tool calls. These are the ones that remove a context switch — the user stays in the assistant instead of opening Findity.

Submit this receipt as a travel expense on my Berlin trip report.

What's the status of my March expense report?

Which expenses are waiting for my approval?

How much have I spent on client entertainment this quarter?

Agent-assisted work

Longer sequences where the assistant does the repetitive part and the user stays in the loop on the decisions.

Bulk capture. The user uploads a batch of receipts at once. The agent categorises each one, drafts a description, assembles them into an expense, and submits it for approval — turning an evening of item-by-item entry into one review pass.

Approver triage. An approver has the agent work through their own approval queue, surfacing which items sit below threshold and within policy, and approving those on the user's instruction. The approvals are the user's own, made with the user's own rights.

Troubleshooting

Next steps