Allow an artificial intelligence to read, write, compile and correct your LaTeX files directly in your Sarmate drive. This guide details every step to connect your AI to your Sarmate workspace.
A so-called "agentic" artificial intelligence is a program installed on your computer, able to carry out tasks autonomously rather than merely responding. It is most often operated from a terminal — the command-input window — though some can be used from a browser or a dedicated application. MCP (Model Context Protocol) is a standard that tells artificial intelligences how to connect to an external service and interact with it. Your AI thus connects to Sarmate's MCP server, which authorises it to read, edit and compile the files in your drive, based on your natural-language instructions. Four notions summarise how it operates.
Installed on your machine, the AI opens, edits and compiles your files autonomously, from requests you formulate in natural language.
The terminal is a window on your computer in which you type your requests: this is where you converse with the AI. (Some AIs also accept the browser.)
When you ask the AI to perform an action from your computer, it carries it out on your sarmate.net drive. And if the file concerned is open in the sarmate.net editor, you see the changes live.
Every change keeps a restorable backup for 30 days, file deletion is disabled, and you can revoke your AI's access to sarmate.net at any time, from the website.
The principle is always the same, whatever your AI:
From sarmate.net, generate a personal key that lets your AI access your drive.
Install the AI of your choice on your computer: Gemini CLI (free), Claude Code… (Le Chat, however, works in the browser, with no installation.)
On your computer, paste the configuration (with your token) into a terminal prompt, then ask the AI, in natural language, to act on your sarmate.net files. (Le Chat, however, is configured in the browser.)
smt_). ⚠️ It is shown only once — keep it for step 3.To begin, you can for example try Gemini CLI, which offers a free agentic-AI tier. If you prefer to avoid the terminal, Le Chat is set up in the browser.
Gemini CLI is a command-line tool that drives an agentic AI from the terminal. A free tier is available.
These tools need Node.js (free software). Choose your system:
nodejs.org and download the LTS version. Run the .msi file and click "Next" all the way.PowerShell, Enter.node --versionnodejs.org, download the LTS version (.pkg file) and install it.Terminal, Enter.node --versionnode --versionmacOS / Linux: if a permission error appears, add sudo at the start of the command.
smt_VOTRE_TOKEN with the token you copied in step 1.geminiIf you wish to avoid the terminal: Le Chat is configured entirely in the browser, in a few clicks. Available on all Le Chat plans.
chat.mistral.ai → left sidebar → Intelligence → Connectors → + Add Connector.Sarmatehttps://mcp.sarmate.net/mcpAuthorization, type Bearer, value smt_VOTRE_TOKENsmt_VOTRE_TOKEN with the token you copied in step 1.The most accomplished agentic AI for code and LaTeX. It requires a Claude subscription (no free version) and is the reference in the field.
These tools need Node.js (free software). Choose your system:
nodejs.org and download the LTS version. Run the .msi file and click "Next" all the way.PowerShell, Enter.node --versionnodejs.org, download the LTS version (.pkg file) and install it.Terminal, Enter.node --versionnode --versionsmt_VOTRE_TOKEN with the token you copied in step 1.claude mcp list → sarmate ✓ Connectedclaude and ask it to act on your Sarmate files.The Claude desktop app. No daily terminal, but a small technical setup at first (and Node.js installed).
These tools need Node.js (free software). Choose your system:
nodejs.org and download the LTS version. Run the .msi file and click "Next" all the way.PowerShell, Enter.node --versionnodejs.org, download the LTS version (.pkg file) and install it.Terminal, Enter.node --versionnode --version~/Library/Application Support/Claude/claude_desktop_config.json%APPDATA%\Claude\claude_desktop_config.jsonsmt_VOTRE_TOKEN with the token you copied in step 1.In the browser, no terminal, but file writing is often limited depending on your OpenAI plan.
https://mcp.sarmate.net/mcp, transport Streamable HTTP, header Authorization: Bearer smt_VOTRE_TOKEN. smt_VOTRE_TOKEN with the token you copied in step 1.AI code editor. No terminal: one config file.
~/.cursor/mcp.json and paste:smt_VOTRE_TOKEN with the token you copied in step 1.Cline : Cline sidebar → MCP Servers → Configure MCP Servers, then paste:
Continue : ~/.continue/config.yaml
smt_VOTRE_TOKEN with the token you copied in step 1.Any MCP-compatible client works with these 3 details:
https://mcp.sarmate.net/mcpAuthorization: Bearer smt_VOTRE_TOKENFor a stdio-only client (no HTTP), use the bridge:
smt_VOTRE_TOKEN with the token you copied in step 1.Bearer ? (3) for Gemini/Claude, does node --version respond (Node.js installed)?Create your token, then follow Gemini CLI (free) in step 2.