Mistral AI MCP server
52 operations0 webhooks
Mistral AI's model API. Run chat, fill-in-the-middle and agent completions, create embeddings, moderate and classify text, read documents with OCR, transcribe and synthesize speech, and manage models, files, batch jobs and agents.
Try Mistral AI with your agent
Connect Mistral AI, then paste this into Claude, ChatGPT or Cursor. For ideas from other apps, see the use cases.
Using Corsair, show me what you can do in Mistral AI on my account, then run one read-only example and show me the result.
Mistral AI tools and triggers
Tools are what your agent can do in Mistral AI through Corsair's MCP server. Triggers are the Mistral AI events a routine can start from.
Agents
- CompleteDestructive
Run an agent on a list of messages with the agent's instructions and tools; spends Mistral credits
agents.complete
- CreateWrite
Create an agent: a model with instructions, tools and default settings (beta). Mistral refuses agent creation for some organizations with 403 "Use Managed Agents instead"; existing agents can still be listed, read, versioned and deleted
agents.create
- DeleteDestructive
Delete an agent; cannot be undone (beta)
agents.delete
- Delete AliasDestructive
Remove a version alias from an agent; the version itself stays (beta)
agents.deleteAlias
- GetRead-only
Get an agent, at its current version or at a given version or alias (beta)
agents.get
- Get VersionRead-only
Get one version of an agent by number or alias (beta)
agents.getVersion
- ListRead-only
List agents, filtered by name, search text or metadata; page with page_token (beta)
agents.list
- List AliasesRead-only
List an agent's version aliases (beta)
agents.listAliases
- List VersionsRead-only
List an agent's versions; page with page and page_size (beta)
agents.listVersions
- Set VersionWrite
Switch an agent to another of its versions (beta)
agents.setVersion
- UpdateWrite
Change an agent; each update creates a new version (beta)
agents.update
- Upsert AliasWrite
Point a version alias such as production at an agent version, creating it if needed (beta)
agents.upsertAlias
Audio Agents
- CreateWrite
Create a voice agent: speech-to-text, LLM and text-to-speech settings, instructions and tools
audioAgents.create
- DeleteDestructive
Delete a voice agent; cannot be undone
audioAgents.delete
- GetRead-only
Get a voice agent, optionally at a given version
audioAgents.get
- ListRead-only
List voice agents; page with page_token
audioAgents.list
- List SessionsRead-only
List a voice agent's sessions; page with page_token
audioAgents.listSessions
- List VersionsRead-only
List a voice agent's versions; page with page_token
audioAgents.listVersions
- UpdateWrite
Replace a voice agent's name, description and configuration
audioAgents.update
Models
- ArchiveWrite
Archive a fine-tuned model; models.unarchive restores it
models.archive
- DeleteDestructive
Delete a fine-tuned model; cannot be undone
models.delete
- GetRead-only
Get a model's details and capabilities
models.get
- ListRead-only
List the base and fine-tuned models this key can use, with their capabilities
models.list
- UnarchiveWrite
Restore an archived fine-tuned model
models.unarchive
- UpdateWrite
Rename a fine-tuned model or change its description
models.update
Batch
- CancelDestructive
Ask Mistral to cancel a batch job: status goes to CANCELLATION_REQUESTED, then CANCELLED; there is no way to resume it
batch.cancel
- CreateDestructive
Start a batch job that runs many requests to one endpoint, from uploaded .jsonl files or inline; spends Mistral credits and needs a Mistral plan with billing enabled (402 otherwise). Follow it with batch.get until status is SUCCESS, FAILED, TIMEOUT_EXCEEDED or CANCELLED
batch.create
- DeleteDestructive
Ask Mistral to delete a batch job; cannot be undone
batch.delete
- GetRead-only
Where a batch job stands: status, request counts, and output_file once it's done
batch.get
- ListRead-only
List batch jobs, filtered by status, model or date; page with page and page_size
batch.list
Files
- DeleteDestructive
Delete an uploaded file; cannot be undone
files.delete
- GetRead-only
Get a file's details
files.get
- Get Signed UrlRead-only
A temporary download URL for a file's content, such as a batch job's output_file
files.getSignedUrl
- ListRead-only
List uploaded files, filtered by purpose or name; page with page and page_size
files.list
- UploadWrite
Upload a file (base64, up to 512 MB) for batch jobs, fine-tuning or OCR; returns its id
files.upload
Voices
- CreateWrite
Create a custom voice from a base64 audio sample
voices.create
- DeleteDestructive
Delete a custom voice; cannot be undone
voices.delete
- GetRead-only
Get a voice
voices.get
- ListRead-only
List preset and custom voices for audio.speech; page with page_token
voices.list
- UpdateWrite
Change a custom voice's name, languages, tags or details
voices.update
Classifiers
- ClassifyDestructive
Classify texts with a fine-tuned classifier model; spends Mistral credits
classifiers.classify
- Classify ChatDestructive
Classify conversations with a fine-tuned classifier model; spends Mistral credits
classifiers.classifyChat
- ModerateDestructive
Score texts against moderation categories (sexual, hate, violence, self-harm…); spends Mistral credits
classifiers.moderate
- Moderate ChatDestructive
Score a conversation against moderation categories; spends Mistral credits
classifiers.moderateChat
Audio
- SpeechDestructive
Turn text into speech; returns base64 audio (mp3 by default); spends Mistral credits
audio.speech
- TranscribeDestructive
Transcribe audio from base64 content, a URL or an uploaded file, optionally with speakers and timestamps; spends Mistral credits
audio.transcribe
Responses
- Count Input TokensRead-only
Count the input tokens a responses.create call would use, without running it
responses.countInputTokens
- CreateDestructive
Generate a response with the Responses API (input items, tools, reasoning); spends Mistral credits
responses.create
Chat
- CompleteDestructive
Generate a chat answer from a list of messages, with optional tools and JSON output; spends Mistral credits
chat.complete
Embeddings
- CreateDestructive
Turn one or more texts into embedding vectors; spends Mistral credits
embeddings.create
Fim
- CompleteDestructive
Fill in code between a prompt and a suffix with a Codestral model; spends Mistral credits
fim.complete
Ocr
- ProcessDestructive
Read a PDF or image (URL or uploaded file) into markdown per page, with optional structured annotations; spends Mistral credits
ocr.process
Connect Mistral AI to Claude, ChatGPT or Cursor
- 1
Connect Mistral AI in Corsair
Add your Mistral AI API key once in Corsair. Corsair stores the credential encrypted and makes the Mistral AI API calls for your agent, so it never goes into a prompt or a config file. Connect Mistral AI
If your link lets the agent add apps, you can skip this: it gives you a sign-in link the first time it needs Mistral AI. On a ChatGPT personal plan, connect Mistral AI here first.
- 2
Add Corsair's MCP server to your agent
Every client uses the same address:
https://mcp.corsair.dev/mcpOpen Customize → Connectors → Add custom connector, paste the address, then select Connect.Add a custom MCP server with the address and choose OAuth. Personal plans get read tools only.Add to CursorOr add this to~/.cursor/mcp.json. Cursor opens Corsair to sign in the first time it connects.{ "mcpServers": { "corsair": { "url": "https://mcp.corsair.dev/mcp" } } }Run this, then run/mcpin Claude Code to sign in.claude mcp add --transport http corsair https://mcp.corsair.dev/mcpCodex, Gemini CLI, VS Code, Zed and other clients with OAuth sign-in work the same way. Clients without it, such as GitHub Copilot's coding agent, use a link's own address and an API key. Setup for every client
- 3
Sign in and choose what needs your approval
Your client opens Corsair to sign in. Pick a link: the apps your agent can use and a mode that decides what it may do without asking. In the default Cautious mode, Mistral AI's read and write operations run right away and its 21 destructive operations wait for your approval. Strict mode asks before every write and refuses destructive operations, and Read mode allows reads only.
FAQ
Install corsair and @corsair-dev/mistralai, add the Mistral AI plugin, and connect with an API key. Then call methods like corsair.mistralai.api.agents.complete() from your app. Corsair keeps each tenant isolated.
Yes. Mistral AI requires an API key from your Mistral AI account or developer dashboard. Create credentials with access that matches what you automate, then paste them into Corsair or connect through managed auth in Hub.
Yes. One Corsair client can hit Mistral AI, GitHub, Notion, Stripe, and the rest. Trigger Mistral AI from a webhook, push updates to Slack, or chain actions across plugins. Same tenant context across integrations, plain TypeScript.
Teams use Corsair for mistral AI's model API. Run chat, fill-in-the-middle and agent completions, create embeddings, moderate and classify text, read documents with OCR, transcribe and synthesize speech, and manage models, files, batch jobs and agents. Common builds include sync jobs, admin tooling, and agents that call Mistral AI APIs from your backend.
The SDK is open source. Self-host and you pay Corsair nothing for the integration layer. Hub bills for hosted auth and execution, not per Mistral AI API call. Hub has a generous free tier. For hobby projects you can probably run the whole thing on free. Paid plans start to matter with more tenants or heavier traffic.
We maintain @corsair-dev/mistralai in the open repo. When Mistral AI ships API changes, upgrade the package. Your call sites stay typed. You should not need to rewrite the integration layer each time.
No. Corsair does not read your Mistral AI data or hold individual access credentials. Credentials stay in your database. API calls run on your instance. We do not store Mistral AI data on your behalf.
Yes. Add https://mcp.corsair.dev/mcp as a custom connector or MCP server in Claude, ChatGPT, Cursor, VS Code, Claude Code, Codex or any client that supports remote MCP servers with OAuth sign-in, then sign in to Corsair. On ChatGPT personal plans, custom servers currently get read tools only.
Yes. Put Mistral AI on a link in Strict mode and every write waits for your approval, while destructive actions are refused. The default Cautious mode asks only before the 21 destructive Mistral AI operations. You can change a link's mode at hub.corsair.dev/me.