AI and MCP

How to connect LM Studio to SpendixAI with MCP

Edit mcp.json in LM Studio to add SpendixAI, sign in through the browser and approve each tool call. Needs LM Studio 0.4.10 or later; begin with read-only access.

Editorial illustration for How to connect LM Studio to SpendixAI with MCP

Running a model on your own computer feels private, and LM Studio makes that easy. This guide adds one outside source to it, your expense records, so the local model can answer questions about spending without you copying figures into the chat.

What stays local and what does not

The model runs on your machine. The records do not live there: they stay in your workspace at SpendixAI, the expense tracker we build for India, and LM Studio fetches them after you sign in. So the sentence to remember is this: the model is local, but the figures it reads come from SpendixAI. Grant only the access you need. The quality of the answers depends on the model you have loaded.

Version check

Remote MCP servers arrived in LM Studio 0.3.17, and the vendor's changelog says OAuth support was added in 0.4.10, with later fixes in 0.4.12 and 0.4.14. Use 0.4.10 at the very least, and a newer release if you can.

Editing mcp.json

  1. In SpendixAI open Developer → MCP connections and copy the server address.
  2. In LM Studio open the Program tab in the right sidebar.
  3. Choose Install, then Edit mcp.json.
  4. Add an entry for spendixai that has a url field.
  5. Save the file. A browser window opens for authorisation.
  6. Sign in to SpendixAI, pick the workspace and choose a permission preset.

The whole entry is a name and a url, where the url is the value this guide calls <your SpendixAI server address>. It sits inside the mcpServers object of the file. LM Studio can also take an auth block with a client ID and secret for servers that do not register clients automatically, but SpendixAI does, so leave it out. No API key is involved, and LM Studio stores the token after you sign in.

Tool confirmations

Before a tool runs, LM Studio shows a confirmation dialog. You can read and edit the arguments, then allow once or always allow. For a finance source, allow once is the careful choice. You can change the saved choices later under App Settings, in Tools and Integrations. The vendor also warns never to install MCP servers from sources you do not trust.

If the browser window never opens, check that the JSON in mcp.json is valid. A missing comma or brace can stop an entry from loading. Edit one entry at a time, save, and watch for the sign-in. If you already use other MCP servers, each has its own entry and its own confirmation behaviour, so changing one does not affect the rest.

When you ask about many transactions, request a summary rather than a full list, for example totals by category for a month. Short answers are easier for a local model to get right, and you can open the SpendixAI report to see the underlying rows.

Summary table

TopicWhat to know
Minimum version0.4.10 for OAuth
Config filemcp.json, opened from the Program tab
EntryName spendixai with a url
Sign-inBrowser window, token stored by LM Studio
ApprovalsDialog before each tool; allow once or always
0.4 menu pathNot confirmed

What the documentation left open

We compared these steps with LM Studio's own documentation in November 2026. The Program, Install and Edit mcp.json path comes from the older 0.3.17 era, and we could not confirm the exact path in the 0.4 interface. If your menu differs, look for the same entry.

Questions to ask

  • What did I spend on food last month compared with the month before?
  • Which merchants did I pay most often this quarter?
  • List my recurring charges and what each costs per year.

Local models can slip on arithmetic, so compare totals with the matching SpendixAI report before you rely on them.

SPENDIXAI JOURNAL

Useful ideas start with a clearer picture of your own spending.

Explore your reports