Mistral AI announced today the release of Connectors in Studio, a new feature that allows developers to build highly customized AI applications grounded in enterprise data. All built-in connectors and custom MCPs are now accessible through API/SDK for use with all model and agent calls. The update also introduces direct tool calling, giving developers precise control over when and how tools are invoked without authentication barriers hindering testing and iteration. Additionally, developers can implement human-in-the-loop approval flows, ensuring secure review and confirmation before tool execution while maintaining flexibility and governance. Programmatic access enables the creation, modification, listing, and deletion of connectors, along with listing their tools and running them directly. All connectors are centrally registered and available across Mistral apps such as LeChat and AI Studio, with Vibe coming soon. Usage via Conversation API, Completions API, and Agent SDK supports complex workflows and integration with enterprise systems like CRMs, knowledgebases, and productivity tools. *Source: [mistral](https://mistral.ai/news/connect-the-dots-build-with-built-in-and-custom-mcps-in-studio)* The release addresses challenges in building enterprise AI agents, such as tracking down API docs, writing and maintaining tool functions, and setting up OAuth. Teams often rebuild the same integration layer, leading to security risks, lack of observability, and duplicated work. A connector packages an integration into a single, reusable entity using the MCP protocol. For example, a custom MCP connector can be created with parameters like name, description, server, visibility, and OAuth configuration. Once registered, the connector is discoverable, governed, and monitored in Studio, becoming a native tool for conversations, agents, or workflows without rewriting integration logic or duplicating authentication. Attaching a connector to a conversation requires just one line of code. *Source: [mistral](https://mistral.ai/news/connect-the-dots-build-with-built-in-and-custom-mcps-in-studio)* The release includes a runnable example for building an agent for a multi-step workflow based on reasoning across sources, leveraging secure connectivity to GitHub, public repo content, and live web data. The agent can understand intent, analyze code, and propose changes alongside other use cases like generating tests, refactoring, and identifying vulnerabilities. To use the feature, developers must install the Mistral SDK and set the MISTRAL_API_KEY environment variable. The example demonstrates creating a custom MCP connector for querying code bases via the DeepWiki remote server, which provides an MCP interface to API/tool endpoints. This allows the agent to explore content and documentation without manual scraping or loading entire repos. The MCP server is registered once, enabling reuse across conversations, agents, or direct tool calls. *Source: [mistral](https://mistral.ai/news/connect-the-dots-build-with-built-in-and-custom-mcps-in-studio)*