--- title: 'AI Load Testing With a French LLM: OctoPerf MCP Meets Mistral Vibe' url: https://blog.octoperf.com/ai-load-testing-with-a-french-llm-octoperf-mcp-meets-mistral-vibe/ date: '2026-06-29' authors: - gpereira categories: - Innovation tags: - Load Testing - Performance Testing - Artificial Intelligence - On Premise description: Drive your OctoPerf load tests in plain language from Mistral Vibe, the French LLM, using a Custom MCP Connector. A hands-on guide to connecting Vibe to the OctoPerf MCP Server in a few clicks, with European data sovereignty built in. --- # AI Load Testing With a French LLM: OctoPerf MCP Meets Mistral Vibe When we [released the OctoPerf MCP Server](https://blog.octoperf.com/octoperf-mcp-server/), most teams connected to it straight from [Claude.ai](https://claude.ai). Then we showed how to run [the whole stack on-premise](https://blog.octoperf.com/octoperf-mcp-server-fully-on-premise-ai-load-testing-with-a-local-llm/) with a local model. But a question kept coming back from European teams: **can we drive our load tests with a French LLM, hosted in Europe, instead of a US model?** The answer is yes, and it takes about five minutes. [Mistral Vibe](https://mistral.ai/products/vibe/), the assistant **formerly known as Le Chat**, now supports **custom Model Context Protocol connectors**, so you can point it straight at the OctoPerf MCP Server and start testing in plain French (or English). ## Table of Contents - [Why a French LLM Changes the Equation](#why-mistral) - [What You Need](#prerequisites) - [Step 1 - Grab the OctoPerf MCP Endpoint](#mcp-endpoint) - [Step 2 - Add a Custom MCP Connector in Mistral Vibe](#add-connector) - [Step 3 - Authenticate With OAuth](#authenticate) - [Step 4 - Run a Test Prompt](#test-prompt) - [What About Vibe Code, Mistral's Claude Code?](#vibe-code) - [What This Unlocks](#benefits) - [Conclusion](#conclusion) ## Why a French LLM Changes the Equation { #why-mistral } ![OctoPerf connected to Mistral AI through the Model Context Protocol](https://blog.octoperf.com/img/blog/octoperf-mcp-server-mistral-le-chat/octoperf-mistral-ai-mcp.png "OctoPerf and Mistral AI, bridged by MCP") [Mistral AI](https://mistral.ai/) is a Paris-based lab whose models are **trained and hosted in Europe**, which matters when your prompts carry production URLs, test data or business context. For finance, healthcare and public-sector teams bound by GDPR or data-residency rules, **keeping the LLM inside European jurisdiction** is often the deciding factor between adopting an AI workflow and shelving it. The piece that makes this practical is the [Model Context Protocol](https://modelcontextprotocol.io). MCP is an open standard, so the OctoPerf MCP Server does not care which model calls it: Claude, a local [Qwen3](https://qwenlm.github.io/), or **Mistral Vibe** all speak the same protocol. Mistral shipped its [Custom MCP Connectors](https://docs.mistral.ai/vibe/work/connectors/mcp-connectors) for Vibe alongside its [Agents API](https://mistral.ai/news/agents-api/), which means you can wire **around 100 OctoPerf tools** into a French chat assistant without writing a line of code. ## What You Need { #prerequisites } The setup is deliberately light: - a **Mistral Vibe account**, where you are the administrator (on Free, Pro and Student plans the account owner is admin by default), - an **OctoPerf account** on the [SaaS platform](https://api.octoperf.com/ui/), which already exposes the MCP Server over HTTPS, - nothing to install locally, no API key to generate. One technical requirement is worth flagging up front: Vibe connects to MCP servers over **Streamable HTTP and HTTPS with a valid TLS certificate**. The OctoPerf SaaS endpoint satisfies this out of the box, and we cover the self-hosted case in Step 1. ## Step 1 - Grab the OctoPerf MCP Endpoint { #mcp-endpoint } There is nothing to deploy. The OctoPerf MCP Server runs as a hosted endpoint: ```text https://api.octoperf.com/mcp ``` That single URL is all Vibe needs, keep it handy for the next step. **Prefer your own instance?** The MCP server is **bundled with OctoPerf Enterprise**, so a self-hosted deployment exposes the exact same endpoint under your own domain, `https:///mcp`. Our [fully on-premise guide](https://blog.octoperf.com/octoperf-mcp-server-fully-on-premise-ai-load-testing-with-a-local-llm/) walks through the Docker install end to end. Just remember Vibe runs in the cloud, so that route must be reachable over **public HTTPS with a valid certificate** (a TLS reverse proxy in front of OctoPerf), then everything below works identically against your own server. ## Step 2 - Add a Custom MCP Connector in Mistral Vibe { #add-connector } ![The Connectors entry under Context in the Mistral Vibe Work sidebar](https://blog.octoperf.com/img/blog/octoperf-mcp-server-mistral-le-chat/mistral-ai-connectors-menu.png "Vibe sidebar: Context, then Connectors") Open [Mistral Vibe](https://chat.mistral.ai/work), and in the **Work** sidebar head to **Context ▸ Connectors**, then click **+ Add Connector** and switch to the **Custom MCP Connector** tab. ![Filling in the Custom MCP Connector form with the OctoPerf endpoint](https://blog.octoperf.com/img/blog/octoperf-mcp-server-mistral-le-chat/mistral-ai-create-custom-mcp-connector.png "Title, Server URL, OAuth 2.1 and a Private visibility") Fill in the form, **only the first two fields are mandatory**: - **Title**: a human-friendly name like `OctoPerf MCP`, Vibe derives a reference (`octoperf_mcp`) from it, - **Server**: the endpoint from Step 1, `https://api.octoperf.com/mcp`, - **Description** (optional): a short note like *Load Testing*, - **Authentication Method**: pick **OAuth2.1**, which is what the OctoPerf MCP Server uses, - **Visibility**: leave it **Private** to keep the connector to yourself, or share it with your workspace. Then submit the form. Because you selected [OAuth 2.1](https://oauth.net/2.1/), Vibe prepares the consent flow for the next step rather than asking you for a token. ## Step 3 - Authenticate With OAuth { #authenticate } ![The OctoPerf MCP connector showing a green Connected status in Mistral Vibe](https://blog.octoperf.com/img/blog/octoperf-mcp-server-mistral-le-chat/mistral-ai-octoperf-mcp-connected.png "A green Connected dot confirms the OAuth handshake succeeded") Because the OctoPerf MCP Server uses **OAuth 2.1 with dynamic client registration**, there is **no token to paste**. When Vibe prompts you, it opens the **OctoPerf login** in a guided consent flow. Sign in, approve the access, and the connector flips to a green **Connected** status, with the authorization cached for next time. From that moment, **every tool call runs as you** inside OctoPerf, scoped to your own workspaces and projects, and you can revoke the access anytime from OctoPerf's **Connected Apps** page. No shared secret, no service account, no credentials living in a config file. One honest caveat: Vibe's custom connectors currently focus on **tool calls** and, [per Mistral's own documentation](https://docs.mistral.ai/vibe/work/connectors/mcp-connectors#limitations), do not yet support a few optional MCP capabilities (dynamic tool discovery, resources and prompt templates). That has no impact here, because the OctoPerf workflow is driven entirely by its **tool catalogue**, which Vibe reads at connection time. ## Step 4 - Run a Test Prompt { #test-prompt } ![Mistral Vibe answering a List my OctoPerf workspaces prompt with a table](https://blog.octoperf.com/img/blog/octoperf-mcp-server-mistral-le-chat/mistral-ai-list-octoperf-workspaces.png "Vibe calls list_workspaces and returns a table of workspaces with deep links") Time to prove the chain works. Enable the **OctoPerf MCP** connector in your conversation, then start with a simple discovery prompt (Vibe understands the request in **French or English**): > **You:** List my OctoPerf workspaces. > **Vibe:** *(calls `list_workspaces`)* > > Here are your **5 OctoPerf workspaces**, each with a deep link back to its page: > > 1. **Tests**, > 2. **Default**, > 3. **Crawlers**, > 4. **Default workspace** *(imported from BlazeMeter)*, > 5. **Test analyse**. > > Want me to open one or start a new test? That round trip alone confirms the full chain works: Mistral's model, the MCP connector, and the OctoPerf backend. Now push it into real work, chaining several tools in one request: > **You:** Create a project *Local Checkout*, import the attached HAR file as a Virtual User and run a functional validation. > **Vibe:** *(calls `create_project`, `import_har_virtual_user`, then `validate_virtual_user`)* > > Done: > > 1. Created project **Local Checkout**, > 2. Imported `checkout.har` as a Virtual User with 24 actions, > 3. Launched a validation. **9 of 24 actions are failing**, mostly authentication errors that look like a missing correlation. Want me to triage them and auto-correlate? This is the same [validation-triage and auto-correlation workflow](https://blog.octoperf.com/octoperf-mcp-server/) we showed with Claude, except **the model answering you is European**. The OctoPerf side is identical, only the brain changed. ## What About Vibe Code, Mistral's Claude Code? { #vibe-code } The chat assistant is only half of [Mistral Vibe](https://mistral.ai/products/vibe/). The other half is **[Vibe Code](https://mistral.ai/products/vibe/code/)**, an **agentic coding tool** that is Mistral's direct answer to [Claude Code](https://code.claude.com/docs/en/overview): a terminal CLI (`vibe`), plus VS Code, JetBrains and Zed extensions, powered by the [Devstral](https://mistral.ai/news/devstral/) model. It is **open source** (Apache 2.0, the [`mistralai/mistral-vibe`](https://github.com/mistralai/mistral-vibe) repository), so you can read and self-host it. Like the chat, **Vibe Code speaks MCP**. You declare servers in a [`config.toml` file](https://docs.mistral.ai/vibe/code/cli/mcp-servers), with `http`, `streamable-http` or `stdio` transports: ```toml [[mcp_servers]] name = "octoperf" transport = "streamable-http" url = "https://api.octoperf.com/mcp" ``` There is one honest catch today. The Vibe Code CLI **does not yet support MCP servers that authenticate with OAuth**, it only handles a static `Authorization` header or an API key. The OctoPerf MCP Server, on the other hand, is **OAuth 2.1 only**: every call carries a JWT minted by OctoPerf's identity provider, and there is no static-token shortcut. So for now, **the OctoPerf connector works from Vibe (the chat), not yet from the Vibe Code CLI**. This is a moving target on both sides: the moment Vibe Code adds OAuth MCP support, the same `config.toml` above will light up your roughly 100 OctoPerf tools straight from the terminal. We will cover that flow in a follow-up once it lands. ## What This Unlocks { #benefits } Pairing OctoPerf with a French LLM is not just a compliance checkbox, though it ticks that one cleanly: - **European data residency**: your prompts and test context are processed by a model hosted in Europe, a frequent hard requirement in regulated sectors, - **zero install**: no local model, no GPU, no Docker, just a connector URL and an OAuth login, - **the same MCP tools and skills**: identical [bundled skills](https://blog.octoperf.com/octoperf-mcp-server-in-action-browser-probes-scheduled-runs-and-smarter-reports/) for validation triage, auto-correlation and scenario diagnosis, - **freedom of model**: MCP is an open standard, so you can switch between Vibe, Claude or a local model without touching OctoPerf. It fits the same picture we painted in [bridging open source and enterprise performance testing](https://blog.octoperf.com/why-modern-teams-need-a-bridge-between-open-source-and-enterprise-performance-testing/): a modern, AI-driven workflow that respects where your data is allowed to go. ## Conclusion { #conclusion } With a **Mistral Vibe account** and one **connector URL**, you get the full OctoPerf AI experience driven by a **French, Europe-hosted LLM**, no install and no API key. The OctoPerf MCP Server already speaks the open Model Context Protocol, so adding Vibe is a five-minute, no-code job. If you need everything inside your own walls, our [fully on-premise guide](https://blog.octoperf.com/octoperf-mcp-server-fully-on-premise-ai-load-testing-with-a-local-llm/) runs the same MCP tools against a local model with nothing leaving your network. And if you are still evaluating, the [SaaS platform](https://octoperf.com/) lets you try this exact Mistral-driven workflow in minutes.