Which Kimi we mean
This guide is about Kimi Code CLI, the terminal coding agent from Moonshot AI. It is not the Kimi chat app. The documentation we read describes adding remote MCP servers in the CLI only, and shows no way for a user to add their own server in the chat app. Kimi Code also has a VS Code extension, but the pages we read did not describe its MCP setup, so this guide leaves it out. Everything here is about the terminal tool, which you start from your own shell.
Worth saying up front that we build SpendixAI, so treat our view of it as a vendor’s view.
The setup in four moves
- Copy the server address from Developer → MCP connections in SpendixAI.
- In Kimi Code CLI, run /mcp-config to add an entry, or edit the mcp.json file yourself.
- Run /mcp-config login spendixai and authorise in the browser that opens.
- On the SpendixAI consent screen, sign in, choose a workspace and pick Read only.
What the file looks like
In mcp.json, there is an object called mcpServers. Inside it you add one entry named spendixai, and that entry has a single field, url, set to <your SpendixAI server address>. An HTTP server is described only by its url, with no transport field.
The user-level file sits in the .kimi-code folder of your home directory, or under the folder named by the KIMI_CODE_HOME setting. A project file in .kimi-code overrides it. For personal finance, we would keep the entry at user level so it never travels with a repository.
How tools appear and get approved
Kimi Code CLI names MCP tools with the pattern mcp, then the server name, then the tool name, each joined by double underscores. A call that matches no permission rule prompts you. Choosing Approve for this session allows later calls of that kind. Permanent rules go in config.toml, each with a decision and a pattern.
Confirmed and unconfirmed
| Item | Status |
|---|---|
| Remote HTTP with OAuth | Documented |
| Login command | /mcp-config login, per the vendor page |
| Plan or minimum version | Not stated |
| A kimi mcp add command with an HTTP option | Seen only in secondary results; not used here |
| VS Code extension setup | Not covered by the page we read |
After the login finishes, ask something you can check yourself before you trust any larger answer. A good first test is to compare one total with the matching SpendixAI report.
Sample questions
- 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.
Choosing a preset
SpendixAI's consent screen gives three choices. Read only is for looking at transactions and reports. Read and add lets the agent also create transactions and upload documents. Full access adds categorisation rules and insights. Billing and member settings are never part of it, and an assistant cannot exceed the role you hold in the workspace.
What the answers rest on
SpendixAI holds the statements you uploaded as PDF, CSV or Excel and the expenses you added by hand. It does not connect to your bank or read SMS, so a question about last week only works if that week has been imported. The wording of each answer comes from the model behind Kimi Code CLI, under its provider's terms, which is why SpendixAI suggests checking important figures in its own reports.
Ending the connection
Remove the entry from mcp.json, or delete it in /mcp-config, and then disconnect in SpendixAI under Developer → MCP connections. A member's role in the workspace also caps what any assistant can do, and access is checked again on every call.
Moonshot AI's documentation was our source, and we last checked it in November 2026. If a command differs from what you see, follow the vendor's current guide.
Useful ideas start with a clearer picture of your own spending.
Explore your reports

