MCP
The Mistral MCP server connects an AI agent to your Studio workspace over the Model Context Protocol. It exposes your workspace as MCP tools and authenticates each request with your Mistral API key.
Use it from wherever your agent runs: a coding CLI, an IDE, a desktop assistant, or your own application.
| Endpoint | https://api.mistral.ai/mcp |
| Transport | Streamable HTTP |
| Authentication | Authorization: Bearer <MISTRAL_API_KEY> |
Before you begin
- A Mistral API key. Keys are workspace-scoped, so the agent gets the access that key already has.
- An MCP client that supports the streamable HTTP transport and static request headers.
Export the key in the environment that launches your client:
export MISTRAL_API_KEY="your_api_key"Connect a client
vibe mcp add mistralai \
--url https://api.mistral.ai/mcp \
--transport streamable-http \
--api-key-env MISTRAL_API_KEYThis writes the entry to ~/.vibe/config.toml:
[[mcp_servers]]
name = "mistralai"
transport = "streamable-http"
url = "https://api.mistral.ai/mcp"
[mcp_servers.auth]
type = "static"
api_key_env = "MISTRAL_API_KEY"The configuration stores the variable name and the environment holds the key. Run /mcp mistralai in a session to list the tools. See MCP servers for permissions and tool filtering.
What the server exposes
The server groups its tools by Studio capability.
| Capability | What an agent can do |
|---|---|
| Skills | Browse the same Skill catalog as Studio, read a Skill and its files, write a new version, change sharing, or delete a Skill. |
| Workflows | List workflows, deploy managed workflows, start and signal executions, read events and logs, and manage schedules. |
| Connectors | List your connectors and read the details of each one. |
| Observability | Search traces, spans, and logs from your applications, and open a single trace or span. |
| Evaluations | Manage evaluation projects and evaluations, read runs and records, and follow optimizations. |
| Organization | Look up workspace members and resolve a name to a user ID. |
| Apps | List your apps, read an app's modules, environments, and running versions, read deploy runs and module or gateway logs, redeploy or roll back a module to a version it already ran, and publish or unpublish an app in the catalog. Env var values are never returned, only their names. |
Observability, Evaluations, and Apps tools appear only for workspaces where those features are turned on. The server lists the tools available to your key, so ask your client for its tool list, or call tools/list. Each tool describes its own arguments.
Access and permissions
Every call uses your API key's permissions.
Some tools write. They add Skill versions, change sharing, delete Skills, deploy workflows, start or terminate executions, change schedules, create or delete evaluations, redeploy or roll back app modules, and publish or unpublish apps. Redeploying an app module never ships new code. Use mistral apps deploy for that. Configure per-tool approval in your client so you can review those calls before they run.
Troubleshooting
| Problem | Cause | Solution |
|---|---|---|
401 {"detail":"Invalid API Key"} | The key is missing, misspelled, or revoked. | Export the variable in the environment that launches your client, and check the key is active in the console. |
| A tool call returns a permission error. | The operation requires additional workspace access. | Ask a workspace admin for the matching access, then retry. |
| Observability, Evaluations, or Apps tools are missing from the tool list. | The feature is off for your workspace. | Check that the feature is available in Studio, then reconnect the client. |
Next steps
- Create a Skill in Studio to give the agent something to work with.
- Build a workflow that the agent can deploy and run.
- MCP servers in the Vibe Code CLI for per-tool permissions and filtering.
- API keys to create and rotate the key you connect with.