Skip to main content

Connect with Le Chat

This guide walks through connecting Alien Intelligence MCP connectors to Mistral's Le Chat. Le Chat supports custom MCP server connections through its Connectors feature, and provides the richest agent wrapper of the major AI platforms — combining connectors, instructions, tone, guardrails, and knowledge (libraries) into a single agent configuration.

For general MCP concepts and available tools, see AI Agent Integration. For MCP server deployment, see Deploy MCP. For usage examples and expected output, see Connector Examples.

Prerequisites​

  • A deployed MCP server on your data cluster (see Deploy MCP), or a public MCP server URL for open-science connectors (see Alien Marketplace).
  • A Le Chat account with access to the Connectors feature.
  • Your MCP server URL (e.g., https://openaire.mcp.alpha.alien.club/mcp).

Connect via Le Chat​

This section walks through the full setup: adding a custom MCP connector, creating a library for skill-like behavior, and building an agent that ties everything together.

Add a Custom MCP Connector​

  1. Open chat.mistral.ai.
  2. In the left sidebar, expand the Intelligence section. You will see four entries: Memories, Connectors, Libraries, and Instructions.
  3. Click Connectors to open the My Connectors page.
  4. The page displays Featured Connectors — pre-built integrations for Atlassian, Box, GitHub, Gmail, Google Calendar, Linear, Notion, Outlook, Slack, and Stripe, each with its own Connect button.
  5. Click the + Add Connector button in the top right corner.
  6. A dialog appears with two tabs: Connector Directory (a searchable list of pre-built connectors) and Custom MCP Connector.
  7. Click the Custom MCP Connector tab.
  8. Fill in the form:
    • Connector Name: a short identifier, e.g. openaire-alpha
    • Connector Server: your MCP server URL, e.g. https://openaire.mcp.alpha.alien.club/mcp
    • Description: optional, e.g. "Alien Intelligence — OpenAIRE research data"
    • Authentication Method: leave as Auto-detect (the default). Le Chat will auto-detect OAuth2.1 from the server's /.well-known/oauth-authorization-server endpoint.
  9. Note the message: "This connector will be shared within your organization" — custom connectors are org-scoped by default.
  10. Click Create. The form updates to show the detected authentication method (OAuth2.1) with an expandable OAuth2 Advanced Settings section.
  11. Click Connect. Your browser redirects to the OAuth provider (Alien Auth / Authentik).
  12. Log in with your Alien Intelligence credentials (email/password, Google, or GitHub) and authorize Le Chat.
  13. After authentication you are redirected back to the Connectors page. Your connector now appears at the bottom under a "Connectors" section with a green Connected indicator.

Create a Library (Skill Workaround)​

Le Chat does not have a direct "skills" concept, but Libraries serve the same purpose. Upload Markdown files that describe workflows, prompt templates, or domain knowledge, and the agent can search and reference them at runtime.

  1. In the sidebar under Intelligence, click Libraries (or go to chat.mistral.ai/libraries).
  2. The Libraries page lists existing libraries with file counts, sizes, and dates.
  3. Click + New Library.
  4. A new library page opens with an editable name field, buttons for + Upload, Webpage, Share, a three-dot menu, and a drag-and-drop area for files.
  5. Click the name field and rename the library, e.g. OpenAIRE Alpha Skill.
  6. Click + Upload or drag files directly onto the page. An Upload Documents modal opens.
  7. Drop or browse for .md files (up to 100 files, 100 MB per file).
  8. Uploaded files appear in the library's file list showing filename, type, size, upload date, and who added them.

Create an Agent with Connector and Library​

  1. Click Agents in the sidebar (or go to chat.mistral.ai/agents).
  2. Click Create an Agent in the top right.
  3. The agent editor has these sections:
    • Name and Description fields at the top.
    • Instructions area — type natural-language instructions or use the / command to quickly attach tools, connectors, and libraries inline.
    • Add guardrails — optional safety constraints.
    • Tone — tabs for Default, Personality (16 presets: Polite, Casual, Objective, Concise, etc.), and Custom.
    • Knowledge — expandable section containing:
      • Tools: Code Interpreter, Image generation, Canvas, Web search (each toggleable).
      • Connectors: your custom MCP connectors (toggleable checkboxes). Built-in connectors (GitHub, Notion, Slack, etc.) appear alongside custom ones.
      • Libraries: searchable list with checkboxes for each library.
  4. Attach the connector — either type / in the Instructions area and select the connector from the popup, or scroll down to Knowledge > Connectors and check its box.
  5. Attach the library — either type / in Instructions and select the library, or find it under Knowledge > Libraries and check its box.
  6. Attached items appear as chips in the Instructions area: connectors show a blue icon, libraries show a pink icon.
  7. The agent auto-saves. A Saved status appears at the top.
  8. Visibility defaults to Private. Change to Shared if others in your organization should see it.
  9. To use the agent, type @agent-name in any Le Chat conversation.
tip

The / command in the Instructions area is the fastest way to wire up tools, connectors, and libraries. You do not need to scroll through the Knowledge section.

What You See After Setup​

  • The connector appears under Intelligence > Connectors with a green "Connected" dot.
  • The library appears under Intelligence > Libraries with its file count and total size.
  • The agent appears under Agents and can be invoked with @agent-name in any chat.
  • When the agent calls MCP tools, Le Chat shows the tool invocation and results inline in the conversation. Tool annotations (readOnlyHint, title) are not visibly surfaced in Le Chat's UI.
info

There is currently no public submission process for Le Chat's Connectors Directory. All Alien Intelligence connectors use the Custom MCP Connector route described above.

Platform-Managed MCP Servers​

If you have your own data cluster deployed through Alien Intelligence, the MCP server URL is available in the platform dashboard:

  1. In the Alien Intelligence dashboard, navigate to MCP in the left sidebar.
  2. Expand the row for your data cluster.
  3. Copy the Server URL from the connection card.

Use this URL in the Connector Server field when adding a custom connector (see above). The rest of the setup — OAuth, libraries, agents — is identical.

OAuth Authentication​

Le Chat uses OAuth 2.1 to authenticate with Alien Intelligence MCP servers. The flow works as follows:

Key points:

  • OAuth is required. Le Chat's connector system authenticates via OAuth. The MCP server's standard OAuth discovery endpoint handles the handshake automatically.
  • Auto-detect works. Leave the authentication method set to "Auto-detect" when creating the connector — Le Chat discovers and configures OAuth2.1 from the server.
  • Token refresh is automatic. When your access token expires, Le Chat re-authenticates without manual intervention.
  • Your permissions are enforced. Le Chat can only access data that your Alien Intelligence account is authorized to see.

Using the Agent​

Once your agent is set up with the connector, invoke it with @agent-name and ask questions in natural language. Le Chat selects the appropriate MCP tools automatically.

Example prompts:

  • "Search my documents for research on quantum computing applications."
  • "Get the full text of entry 7890 and highlight the key findings."
  • "Find the most recent documents about climate modeling in my dataset."
  • "List all figures from the paper about neural network architectures."
  • "Find documents conceptually related to 'graph neural networks in drug discovery', even if they use different terminology."

Le Chat will call tools like datacluster_keyword_search, datacluster_vector_search_chunks, datacluster_get_entry_content, datacluster_get_entry_documents, and datacluster_get_entry_file as needed.

Test the full chain:

  1. "List my datasets." — Calls datacluster_list_datasets.
  2. "Search for documents about [topic]." — Calls datacluster_keyword_search.
  3. "Read the content of the top result." — Calls datacluster_get_entry_content.
  4. "Show me the figures from that paper." — Calls datacluster_get_entry_documents and datacluster_get_entry_file.

Troubleshooting​

Connector Does Not Appear After Creation​

  • Refresh the Connectors page. The connector should appear at the bottom under the "Connectors" section.
  • If you see an error during creation, verify that the MCP server URL is reachable. Check the health endpoint: https://your-cluster.mcp.alien.club/health.

OAuth Login Fails​

  • Verify your MCP server is deployed and running.
  • Ensure your data cluster is in an Active state in the platform dashboard.
  • Check that your browser allows redirects and popups from chat.mistral.ai.
  • If the auto-detect step fails, expand OAuth2 Advanced Settings in the connector form and verify the discovered endpoints.

Connector Shows "Disconnected"​

  • The MCP server may have been redeployed. Click the connector and re-authenticate by clicking Connect again.
  • Try removing and re-adding the connector.
  • Wait a minute between attempts — the server may still be provisioning after a restart.

Agent Does Not Call the Tools​

  • Verify the connector checkbox is enabled in the agent's Knowledge > Connectors section.
  • Check that the connector shows a green "Connected" dot on the Connectors page.
  • Try being explicit: "Using my Alien data connector, search for documents about..."
  • If the connector was recently added, try starting a new conversation and invoking the agent with @agent-name again.

Results Are Empty​

  • Confirm your data cluster has at least one dataset with processed entries.
  • Try a broad search term first to verify connectivity, then narrow down.
  • Check that your account has Viewer (or higher) permissions in the organization that owns the data cluster.

Library Files Not Referenced by Agent​

  • Verify the library is checked in the agent's Knowledge > Libraries section, or attached via the / command in Instructions.
  • Ensure uploaded files are in .md format and not empty.
  • Try asking the agent directly: "What do you know from the [library name] library?"

Next Steps​