Mistral AI connector for AI agents

Tools your document agent reaches for on Mistral AI, scoped per user.

CALL ANY TOOL

32 tools for LLM API calls in Mistral AI: chat, agent, and code completions, stored conversations, document OCR, embeddings, moderation, fine-tuned classifiers, batch jobs, agents, and workspace files.

mistral_append_to_conversation

Append to conversation

Add new inputs to an existing Mistral conversation and run the next reply on the full history. Returns the conversation_id, the new output entries (assistant messages, tool executions or function calls) and token usage. Use this for the next turn, or to send a function.result after a function call. Use mistral_restart_conversation to branch from an earlier entry.

Parameters

Name

Type

Required

Description

conversation_id

string

Required

ID of the conversation to continue, from mistral_start_conversation or mistral_list_conversations. Example: conv_0195b2f2a6f97c3bb6b1c0f5e1a2d3c4.

inputs

string

Required

What to add to the conversation: a plain string is sent as one user message. For more control pass an array of entries, such as [{"type": "message.input", "role": "user", "content": "..."}], or, to answer a function call the model made, [{"type": "function.result", "tool_call_id": "<id from the function.call entry>", "result": "<your result as a string>"}]. Example: "Summarise the key points of our last discussion."

completion_args

object

Optional

Default generation settings for the agent or conversation: temperature, top_p, max_tokens, random_seed, stop, presence_penalty, frequency_penalty, tool_choice (auto, none, any or required), reasoning_effort and response_format (same shapes as in chat completions). Example: {"temperature": 0.3, "max_tokens": 1024}.

handoff_execution

string

Optional

With multi-agent handoffs: server lets Mistral run the handoff, client returns it to you to perform. One of: `server`, `client`.

store

boolean

Optional

Whether Mistral stores the new entries. Default true.

mistral_cancel_batch_job

Cancel batch job

mistral_classify_chat

Classify chat

mistral_classify_text

Classify text

mistral_create_agent

Create agent

Create a Mistral agent: a saved model, instructions, tools and generation settings you can reuse in conversations and agent completions. Returns the new agent with its id, version, model, name, instructions, tools, completion_args and timestamps. Use this to set up a reusable assistant. Use mistral_update_agent to change one, and mistral_start_conversation or mistral_create_agent_completion to run it.

Parameters

Name

Type

Required

Description

model

string

Required

Model the agent runs on. Example: mistral-medium-latest.

name

string

Required

Name of the agent. Example: Contract reviewer.

completion_args

object

Optional

description

string

Optional

Short description of what the agent does. Example: Flags risky clauses in contracts.

handoffs

array

Optional

IDs of other agents this agent can hand the conversation to. Example: ["ag_0195b2f2a6f97c3bb6b1c0f5e1a2d3c5"].

instructions

string

Optional

System instructions the agent follows in every conversation. Example: You review contracts and list risky clauses with a short reason for each.

metadata

object

Optional

Your own key-value metadata for the agent. Example: {"team": "legal"}.

tools

array

Optional

Tools the agent can use. Built-in tools: {"type": "web_search"}, {"type": "web_search_premium"}, {"type": "code_interpreter"}, {"type": "image_generation"} and {"type": "document_library", "library_ids": ["<library id>"]}. Your own functions use {"type": "function", "function": {"name": ..., "description": ..., "parameters": <JSON Schema>}}; the conversation then pauses with a function.call entry until you send back a function.result. Example: [{"type": "web_search"}].

version_message

string

Optional

Note describing this version of the agent. Example: Added web search.

mistral_create_agent_completion

Create agent completion

Run a saved Mistral agent on a list of messages and get its next reply, using the agent's model, instructions and settings. Returns the completion id, model, choices (message.content, any message.tool_calls, finish_reason) and token usage, in the same shape as a chat completion; non-streaming only. Use this for a stateless call to an existing agent. Use mistral_start_conversation for a server-stored conversation that runs the agent's built-in tools, and mistral_create_chat_completion to call a model directly.

Parameters

Name

Type

Required

Description

agent_id

string

Required

ID of the agent to run, from mistral_list_agents or Mistral AI Studio. Example: ag_0195b2f2a6f97c3bb6b1c0f5e1a2d3c4.

messages

array

Required

Conversation so far, oldest first. Each message has a role (system, user, assistant or tool) and content. Content is plain text, or an array of chunks so a user message can also carry images ({"type": "image_url", "image_url": "https://... or data:image/png;base64,..."}) or documents ({"type": "document_url", "document_url": "https://..."}) for vision-capable models. To return a tool result, add the assistant message with its tool_calls from the previous response, then a tool message with tool_call_id and content. Example: [{"role": "system", "content": "You are a concise assistant."}, {"role": "user", "content": "Who is the best French painter? Answer in one sentence."}]

max_tokens

integer

Optional

Maximum number of tokens to generate. The prompt plus max_tokens must fit in the model's context length. Leave empty to let the model stop on its own. Example: 512.

random_seed

integer

Optional

Seed for random sampling. The same seed and inputs give repeatable results. Example: 42.

response_format

object

Optional

Output format. {"type": "text"} is the default. {"type": "json_object"} turns on JSON mode: the reply is valid JSON, and you must also ask for JSON in a system or user message. {"type": "json_schema", "json_schema": {"name": "book", "schema": {...}, "strict": true}} makes the reply follow your JSON Schema. Example: {"type": "json_object"}.

stop

string

Optional

Stop generating when the model writes this string, or any string in this array. Example: ["\n\n", "END"].

tool_choice

string

Optional

auto, none, any or required, or {"type": "function", "function": {"name": "..."}} to force one function. Example: auto.

tools

array

Optional

Extra functions the agent may call in this request, in the same shape as chat completion tools. Example: [{"type": "function", "function": {"name": "get_weather", "description": "Get the current weather for a city", "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]}}}]

mistral_create_batch_job

Create batch job

Queue a Mistral batch inference job that runs many requests against one endpoint asynchronously, from inline requests or uploaded .jsonl files. Returns the new job with its id, status (QUEUED at first), endpoint, model and request counts. Use this for large workloads that do not need an immediate answer. Track it with mistral_get_batch_job; stop it with mistral_cancel_batch_job. Usage is billed to the API key's workspace at batch pricing.

Parameters

Name

Type

Required

Description

endpoint

string

Required

API endpoint every request in the batch is sent to: /v1/chat/completions, /v1/embeddings, /v1/fim/completions, /v1/moderations, /v1/chat/moderations, /v1/ocr, /v1/classifications, /v1/chat/classifications, /v1/conversations, /v1/audio/transcriptions. Example: /v1/chat/completions. One of: `/v1/chat/completions`, `/v1/embeddings`, `/v1/fim/completions`, `/v1/moderations`, `/v1/chat/moderations`, `/v1/ocr`, `/v1/classifications`, `/v1/chat/classifications`, `/v1/conversations`, `/v1/audio/transcriptions`.

agent_id

string

Optional

Agent to run every request with, instead of a model. Example: ag_0195b2f2a6f97c3bb6b1c0f5e1a2d3c4.

input_files

array

Optional

IDs of .jsonl files already uploaded with purpose batch, where each line is {"custom_id": ..., "body": {...}}. Provide input_files or requests. Example: ["6a5bd9f5-0d4e-4bd0-a6f2-2d2c3b1e7c11"].

metadata

object

Optional

Your own key-value metadata for the job. Example: {"job_type": "nightly-summaries"}.

model

string

Optional

Model to run every request with. Provide model or agent_id. Example: mistral-small-latest.

requests

array

Optional

Requests to run, given inline. Each item has an optional custom_id and a body with the request payload for the endpoint, without model. Example: [{"custom_id": "0", "body": {"max_tokens": 100, "messages": [{"role": "user", "content": "What is the best French cheese?"}]}}]. Provide requests or input_files.

timeout_hours

integer

Optional

Hours after which unfinished requests time out. Mistral's default is 24. Example: 24.

mistral_create_chat_completion

Create chat completion

Send a conversation to a Mistral model and get the next assistant message, with optional function calling, JSON mode or JSON-schema output, and image or document inputs on vision models. Returns the completion id, model, choices (each with message.content, any message.tool_calls and finish_reason) and token usage; non-streaming only. Use this for a single synchronous answer from a model you name. Use mistral_create_agent_completion to run a saved agent, mistral_start_conversation for server-stored conversations with built-in tools, and mistral_create_fim_completion for code infilling. Usage is billed to the API key's workspace.

Parameters

Name

Type

Required

Description

messages

array

Required

model

string

Required

ID of the Mistral model to run, for example mistral-large-latest, mistral-medium-latest, mistral-small-latest, magistral-medium-latest (reasoning) or a fine-tuned model id. Use mistral_list_models to see what your key can call.

frequency_penalty

number

Optional

From -2 to 2. Higher values discourage repeating words that already appear often in the output. Example: 0.5.

max_tokens

integer

Optional

Maximum number of tokens to generate. The prompt plus max_tokens must fit in the model's context length. Leave empty to let the model stop on its own. Example: 512.

integer

Optional

Number of completions to generate for the request; each is returned as a separate choice. Input tokens are billed once. Example: 2.

parallel_tool_calls

boolean

Optional

Whether the model may call several tools in one turn. Mistral's default is true. Example: false.

prediction

object

Optional

Expected output, which speeds up responses when most of the answer is known in advance, such as small edits to a document or code file. Shape: {"type": "content", "content": "<expected text>"}.

presence_penalty

number

Optional

From -2 to 2. Higher values push the model to use a wider variety of words and topics. Example: 0.5.

prompt_mode

string

Optional

Set to reasoning to use Mistral's reasoning system prompt with reasoning (Magistral) models. Leave empty otherwise. One of: `reasoning`.

random_seed

integer

Optional

Seed for random sampling. The same seed and inputs give repeatable results. Example: 42.

reasoning_effort

string

Optional

How much the model reasons before answering, on models that support adjustable reasoning: none, minimal, low, medium, high or xhigh. Example: medium. One of: `none`, `minimal`, `low`, `medium`, `high`, `xhigh`.

response_format

object

Optional

safe_prompt

boolean

Optional

Prepend Mistral's safety system prompt to the conversation. Default false. Example: true.

service_tier

string

Optional

auto serves the request on priority capacity when your workspace has it; standard_only always uses standard capacity. One of: `auto`, `standard_only`.

stop

string

Optional

Stop generating when the model writes this string, or any string in this array. Example: ["\n\n", "END"].

temperature

number

Optional

Sampling temperature from 0 to 1.5. Mistral recommends 0.0 to 0.7: lower is more focused and deterministic, higher is more random. Change this or top_p, not both. Leave empty for the model's default. Example: 0.3.

tool_choice

string

Optional

How the model uses the tools: auto (default, model decides), none (never call a tool), any or required (must call at least one tool), or {"type": "function", "function": {"name": "get_weather"}} to force one function. Example: auto.

tools

array

Optional

Functions the model may call. Each tool is {"type": "function", "function": {"name", "description", "parameters": <JSON Schema>}}. When the model calls one, the response has finish_reason tool_calls and message.tool_calls; run the function yourself and send the result back as a tool message. Example: [{"type": "function", "function": {"name": "get_weather", "description": "Get the current weather for a city", "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]}}}]

top_p

number

Optional

Nucleus sampling: only tokens within this cumulative probability mass are considered, so 0.1 keeps the top 10%. Change this or temperature, not both. Example: 0.9.

mistral_create_embeddings

Create embeddings

mistral_create_fim_completion

Create fim completion

mistral_delete_agent

Delete agent

mistral_delete_conversation

Delete conversation

mistral_delete_file

Delete file

mistral_get_agent

Get agent

mistral_get_batch_job

Get batch job

mistral_get_conversation

Get conversation

mistral_get_conversation_history

Get conversation history

mistral_get_file

Get file

mistral_get_file_signed_url

Get file signed url

mistral_list_agents

List agents

mistral_list_batch_jobs

List batch jobs

mistral_list_conversation_messages

List conversation messages

mistral_list_conversations

List conversations

mistral_list_files

List files

mistral_list_models

List models

mistral_moderate_chat

Moderate chat

mistral_moderate_text

Moderate text

mistral_ocr_document

Ocr document

Extract text, tables, images and layout from a PDF, office document or image with Mistral OCR, given a public URL, a base64 data URI or an uploaded file ID. Returns one entry per page with markdown text, extracted images (with optional base64), tables, page dimensions and optional confidence scores, plus pages_processed and any document_annotation you asked for. Use this to turn a document into markdown or extract structured fields with a JSON schema. Use mistral_create_chat_completion with a document_url chunk to ask questions about a document instead. Billed per page processed.

Parameters

Name

Type

Required

Description

model

string

Required

OCR model to use. mistral-ocr-latest points to the newest OCR model; pinned versions such as mistral-ocr-2512 are also accepted.

bbox_annotation_format

object

Optional

Describe each extracted image or figure with a JSON object that follows your schema, returned as images[].image_annotation. Shape: {"type": "json_schema", "json_schema": {"name": "figure", "schema": {...}}}. Only json_schema is accepted.

confidence_scores_granularity

string

Optional

Add OCR confidence scores per page, per block or per word. Leave empty for no scores, which keeps the response small. One of: `page`, `block`, `word`.

document_annotation_format

object

Optional

Extract structured data from the whole document into a JSON object that follows your schema, returned as document_annotation. Shape: {"type": "json_schema", "json_schema": {"name": "invoice", "schema": {...}, "strict": true}}. Only json_schema is accepted.

document_annotation_prompt

string

Optional

Instructions that guide the document-level extraction. Needs document_annotation_format. Example: Extract the invoice number, date and total amount.

document_name

string

Optional

Optional file name for a document_url input, returned with the result. Example: invoice-2026-03.pdf.

document_url

string

Optional

Document to read, as a public URL or a base64 data URI. Supports PDF, PPTX, DOCX and other document formats, up to 50 MB and 1,000 pages. Example: https://arxiv.org/pdf/2201.04234 or data:application/pdf;base64,JVBERi0x... Provide exactly one of document_url, image_url or file_id.

extract_footer

boolean

Optional

Move each page's footer out of the markdown into the page's footer field. Default false. Needs OCR 2512 or newer.

extract_header

boolean

Optional

Move each page's header out of the markdown into the page's header field. Default false. Needs OCR 2512 or newer.

file_id

string

Optional

ID of a file already uploaded to Mistral with purpose ocr (see mistral_list_files). Provide exactly one of document_url, image_url or file_id. Example: 6a5bd9f5-0d4e-4bd0-a6f2-2d2c3b1e7c11.

image_limit

integer

Optional

Maximum number of images to extract from the document. Example: 10.

image_min_size

integer

Optional

Only extract images whose height and width are at least this many pixels. Example: 100.

image_url

string

Optional

Image to read, as a public URL or a base64 data URI. Supports PNG, JPEG, AVIF and other image formats. Example: https://example.com/receipt.png or data:image/png;base64,iVBORw0KGgo... Provide exactly one of document_url, image_url or file_id.

include_blocks

boolean

Optional

Return paragraph-level bounding boxes for each content block. Mistral's default is true; set false for a smaller response.

include_image_base64

boolean

Optional

Return each extracted image as a base64 data URI in images[].image_base64. This can make the response very large. Default false.

pages

string

Optional

Pages to process, numbered from 0. A string of numbers and ranges such as "0,2-4", or an array of integers such as [0, 1, 2]. Leave empty to process every page.

table_format

string

Optional

How to return tables: markdown keeps them inline in each page's markdown, html returns them separately in each page's tables with placeholders in the markdown. Leave empty for the model default. Needs OCR 2512 or newer. One of: `markdown`, `html`.

mistral_start_conversation

Start conversation

Start a new server-stored conversation with a Mistral model or agent, send the first inputs and run the reply, including built-in tools such as web search or code interpreter. Returns the conversation_id, the new output entries (assistant messages, tool executions or function calls) and token usage. Use this when you want Mistral to keep the history. Continue with mistral_append_to_conversation, branch with mistral_restart_conversation, or use mistral_create_chat_completion for a stateless call. Usage is billed to the API key's workspace.

Parameters

Name

Type

Required

Description

inputs

string

Required

agent_id

string

Optional

Agent to run the conversation with, instead of a model. Provide agent_id or model. Example: ag_0195b2f2a6f97c3bb6b1c0f5e1a2d3c4.

agent_version

integer

Optional

Version number or alias of the agent to use. Leave empty for the agent's current version. Example: 3.

completion_args

object

Optional

description

string

Optional

Description of the conversation. Example: Review of the ACME master services agreement.

handoff_execution

string

Optional

With multi-agent handoffs: server lets Mistral run the handoff, client returns it to you to perform. One of: `server`, `client`.

instructions

string

Optional

System instructions for a model-based conversation. Example: You are a helpful legal assistant. Answer briefly.

metadata

object

Optional

Your own key-value metadata for the conversation. Example: {"customer_id": "cus_123"}.

model

string

Optional

Model to run the conversation with, when not using an agent. Provide model or agent_id. Example: mistral-medium-latest.

name

string

Optional

Name for the conversation. Example: Contract review - ACME.

store

boolean

Optional

Whether Mistral stores the conversation so you can continue it later. Default true; set false for a one-off exchange.

tools

array

Optional

mistral_update_agent

Update agent

For more tools, view docs.

Build your Agent

Same auth pattern across LangChain, OpenAI, Anthropic, and Google ADK.

import { ScalekitClient } from "@scalekit-sdk/node";
import { createReactAgent } from "@langchain/langgraph/prebuilt";

const sk = new ScalekitClient(env.SCALEKIT_ENV_URL, env.SCALEKIT_CLIENT_ID, env.SCALEKIT_CLIENT_SECRET);

// Mistral AI tools scoped to this user
const { tools } = await sk.tools.listScopedTools("user_123", {
  filter: { connectionNames: ["mistral"], toolNames: [
    "mistral_ocr_document",
    "mistral_create_chat_completion",
    "mistral_create_embeddings"] },
  pageSize: 100,
});

const agent = createReactAgent({ llm, tools });
await agent.invoke({ messages: [{ role: "user", content: "Extract the line items from this invoice and summarize them" }] });

import OpenAI from "openai";
import { ScalekitClient } from "@scalekit-sdk/node";

const sk = new ScalekitClient(env.SCALEKIT_ENV_URL, env.SCALEKIT_CLIENT_ID, env.SCALEKIT_CLIENT_SECRET);
const openai = new OpenAI();

const { tools } = await sk.tools.listScopedTools("user_123", {
  filter: { connectionNames: ["mistral"] }, pageSize: 100,
});

const res = await openai.chat.completions.create({
  model: "gpt-5",
  messages: [{ role: "user", content: "Extract the line items from this invoice and summarize them" }],
  tools,
});

// Execute the tool call with the user's vaulted Mistral AI access token
await sk.tools.executeTool(res.choices[0].message.tool_calls[0], "user_123");

import Anthropic from "@anthropic-ai/sdk";
import { ScalekitClient } from "@scalekit-sdk/node";

const sk = new ScalekitClient(env.SCALEKIT_ENV_URL, env.SCALEKIT_CLIENT_ID, env.SCALEKIT_CLIENT_SECRET);
const anthropic = new Anthropic();

const { tools } = await sk.tools.listScopedTools("user_123", {
  filter: { connectionNames: ["mistral"] }, pageSize: 100,
});

const msg = await anthropic.messages.create({
  model: "claude-sonnet-5",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Extract the line items from this invoice and summarize them" }],
  tools,
});

// Tool call runs with the user's vaulted Mistral AI access token
await sk.tools.executeTool(msg.content, "user_123");

import { Agent } from "@google/adk/agents";
import { ScalekitClient } from "@scalekit-sdk/node";

const sk = new ScalekitClient(env.SCALEKIT_ENV_URL, env.SCALEKIT_CLIENT_ID, env.SCALEKIT_CLIENT_SECRET);

const { tools } = await sk.tools.listScopedTools("user_123", {
  filter: { connectionNames: ["mistral"] }, pageSize: 100,
});

const agent = new Agent({
  name: "document_agent",
  model: "gemini-2.5-pro",
  instruction: "Mistral AI tools scoped to this user",
  tools,
});

await agent.run("Extract the line items from this invoice and summarize them");

Try these prompts

Copy any prompt into your agent. Each maps directly to a Mistral AI tool. Click to copy, paste into your agent, done.

Chat and agents

Copy the prompt

Copied

Summarize this contract in plain language.

Copy the prompt

Copied

Run our support agent on this customer message.

Copy the prompt

Copied

Create an agent that answers HR policy questions with these instructions.

Documents and text

Copy the prompt

Copied

Extract text and tables from this scanned PDF.

Copy the prompt

Copied

Create embeddings for these 200 product descriptions.

Copy the prompt

Copied

Check these 10 user comments against the moderation categories.

Batches and files

Copy the prompt

Copied

Queue a batch job for these 5,000 classification requests.

Copy the prompt

Copied

Show the status of last night's batch job.

Copy the prompt

Copied

List the files in our Mistral workspace.

SEE HOW AUTH WORKS

Each user connects their own Mistral AI access token once, and Scalekit sends it with every call. Access tokens stay vaulted, every call is scope checked, and every action is logged.

Authorize

Your user connects

Mistral AI

once. We tie it to their identity and the meetings they approved — no shared bot account, no org-wide access

Who:

user ‘A’

when:

Once per user

access:

Limited to user

Store

Their

Mistral AI

token lives in a vault scoped to them. User A's meetings are never reachable by an agent acting for user B, even on the same connection

vault:

encrypted

scope:

per-user

tokens:

auto-refreshed

Resolve

When your agent calls a

Mistral AI

tool, we fetch the right token server-side. It never touches your agent, never appears in the LLM context, never shows up in your logs

speed:

~40ms

check:

before every call

seen by:

nobody

Audit

Every

Mistral AI

tool call is logged — who triggered it, which meeting was fetched, what came back. 90 days of history, tied to the user who authorized it

history:

90 days

export:

SIEM-ready

logged:

every call

Test other agents

See the same per-user auth pattern across other AI and document connectors.

Support and Ops Teams

Support ticket automation agent

Fetches new Zendesk tickets, drafts a reply from Notion knowledge base articles, digests what it cannot answer to Slack, and archives the rest, acting as the support agent rather than a shared API key.

GTM and RevOps Teams

Competitive intelligence briefing agent

Scans Gong calls for competitor mentions, matches each one to its Notion battlecard, and DMs every affected rep a single Slack digest per cycle. Every call runs as the PMM who owns the briefing, never a shared bot.

Why Scalekit

Secure your agent's access. Connectors ship in minutes

01.

Shared tokens break per-user analytics

A shared Mistral AI access token looks fine in a demo. In production every completion and OCR request looks like one service account, and you cannot tell which user triggered it. Scalekit resolves the credential of the actual user who triggered the agent, never a shared bot.

// shared key
audit → bot_service_account

// scalekit
audit → user_abc ✓

02.

Authentication is not authorization

03.

Multi-tenancy is architectural

04.

Mistral AI today. Ten connectors tomorrow.

“Our agents act across Salesforce, Gong, Google Drive, and more, on behalf of every customer. Scalekit behind the scenes meant we can keep adding tools without ever rebuilding how credentials or tool calling work.”

Venu Madhav Kattagoni

Head of Engineering / Von

FAQs

Frequently Asked Questions

Does the agent access Mistral AI as the user or through a shared access token?

As the user. Each user connects their own Mistral AI access token once, and Scalekit sends it with every call. Audit logs attribute every completion and agent change to that user, not a shared service account.

Where is the Mistral AI access token stored?

In Scalekit's managed AES-256 token vault, namespaced per tenant. To rotate it, the user reconnects with a new access token. Revocation is a single dashboard action. Credentials never appear in prompts, logs, or LLM context.

Can I limit what the agent does in Mistral AI?

Yes. Filter by tool name in listScopedTools to expose only what you want, for example chat, OCR, and embeddings without mistral_delete_agent or mistral_delete_conversation, which permanently delete an agent or a stored conversation. Scalekit also enforces scope checks before every API call.

What happens when a user revokes Mistral AI access?

The connection is invalidated on the next tool call. Subsequent requests for that user fail closed with a clear error. Other users in the tenant remain unaffected. The event is logged for audit.

Can the agent continue a conversation later?

Yes. mistral_start_conversation stores the conversation on Mistral, mistral_append_to_conversation adds inputs and runs the next reply on the full history, and mistral_restart_conversation branches from an earlier entry.