New AI Protocol Lets Chatbots Use Outside Tools and Data

AI chatbots can now connect to outside tools and data thanks to the new Model Context Protocol. This is a big step from just using their own knowledge.
By Newsroom | Technology, Artificial Intelligence |

The Model Context Protocol (MCP) is emerging as a standardized conduit, akin to a universal adapter, for artificial intelligence applications to interface with the outside world. This open-source framework allows Large Language Models (LLMs) to connect with external data sources, tools, and complex workflows, expanding their operational capacity beyond inherent knowledge.

MCP's architecture involves servers that expose data and tools, and clients that develop applications to connect to these servers. These applications can then run inside AI clients, enabling interactive experiences. Essentially, MCP aims to provide a consistent method for AI systems, such as Claude or ChatGPT, to access information and execute tasks through third-party resources.

MCP Agent Enables LLM Querying of Link-State Graphs | Let's Data Science - 1

Data.gouv.fr Explores Conversational Data Access

The French public data portal, data.gouv.fr, has embarked on an experiment utilizing an MCP server. This initiative, launched on February 25, 2026, is framed as a controlled test to explore new avenues for querying and leveraging public data through conversational interfaces. The ambition extends to potentially enabling the editing and publishing of data via these interfaces, albeit with caution and a focus on "sovereign models." Currently, the data.gouv.fr MCP server is limited to read-only access of open public data, with modifications not yet permitted. Concerns have been raised about the inherent difficulty in auditing such systems.

Read More: New Open Source CRM Twenty Lets Developers Code Custom Systems

Google's Integration and Ecosystem

Google has embraced and contributed to the MCP framework, offering official MCP servers and deployment guidance on Google Cloud. Their repository lists a range of services that can be connected via MCP, including databases like AlloyDB for PostgreSQL and Cloud SQL instances, Google Workspace applications (Docs, Sheets, Gmail), and cloud infrastructure such as Compute Engine and Kubernetes Engine. This suggests a move towards making various Google Cloud resources and services accessible through AI-driven interactions.

MCP Agent Enables LLM Querying of Link-State Graphs | Let's Data Science - 2

MCP in Developer Workflows

MCP's practical application is also being integrated into developer tools. For instance, instances exist where an MCP server connection is configured and verified, allowing tools like Copilot, running within VS Code, to interact with external resources like Hugging Face models in real-time. This involves setting up configuration files, often with server details and API tokens, to facilitate this connection.

Read More: AI Human Feedback Method Faces Scrutiny Over Reward Model Flaws

Conceptual Underpinnings

At its core, MCP establishes a standardized, bidirectional link between AI applications and external entities. This allows LLMs to not only retrieve information but also to perform actions in conjunction with data retrieval. This is often described in the context of Retrieval-Augmented Generation (RAG), where LLMs enhance their capabilities by accessing and processing information from external systems. An MCP client, housed within an MCP host, acts as the intermediary, facilitating communication between the LLM and the MCP server, and identifying available servers.

Frequently Asked Questions

Q: What is the new Model Context Protocol (MCP) for AI?
The Model Context Protocol (MCP) is a new way for AI programs, like ChatGPT, to connect to outside information and tools. It helps AI do more than just use its built-in knowledge.
Q: How is data.gouv.fr using the new AI protocol?
France's data portal, data.gouv.fr, started testing an MCP server on February 25, 2026. They want to see if people can ask questions and get public data using AI chatbots. Right now, you can only read data, not change it.
Q: Is Google using the new AI protocol?
Yes, Google is using and helping with the MCP. They offer ways to connect Google Cloud services, like databases and Google Workspace apps, to AI through MCP.
Q: How will developers use the new AI protocol?
Developers can use MCP to connect AI tools, like Copilot in VS Code, to outside resources such as AI models on Hugging Face. This helps AI work with real-time data and tools.
Q: What does the new AI protocol allow AI to do?
MCP creates a link between AI and outside systems. This lets AI get information and also perform actions. It's like giving AI the ability to use external resources to help it answer questions or complete tasks.