What is MCP (Model Context Protocol)?

What problem does MCP solve?

An AI assistant only works with what is in front of it: what it learnt in training and what you paste into the conversation. Before MCP, connecting it to each system required a bespoke integration for every assistant-system pair. MCP replaces those integrations with a single protocol: whoever exposes a system does so once, as an MCP server, and any application that implements the protocol can use it. The specification describes it as a standardised way of sharing context with language models, exposing tools to them, and composing integrations and workflows^[1].

In watching, the consequence is direct: your AI provides analytical capacity; Innguma provides context.

How does it work? Participants, layers and primitives

MCP follows a client-server architecture with three participants^[2]:

1. Host: the AI application the person uses (a chat assistant, a development environment).
2. Client: the component the host creates to maintain the connection with each server; there is one client per server.
3. Server: the program that offers context and capabilities. It can run locally (stdio transport) or remotely (Streamable HTTP transport).

Messages follow the JSON-RPC 2.0 format. Before using anything, the client asks the server what it offers (for example, with tools/list) and then invokes it (tools/call)^[2]. What a server can offer is organised into three primitives:

The client can also offer capabilities to the server, such as elicitation: the server asks the person for a piece of data or a confirmation during a task^[1].

Who maintains the standard, and which version is current?

• November 2024: Anthropic publishes MCP as an open standard^[3].
• December 2025: Anthropic hands it over to the Agentic AI Foundation, a Linux Foundation fund co-founded with Block and OpenAI and supported by Google, Microsoft and AWS, among others. The announcement cites ChatGPT, Gemini and Microsoft Copilot among the products that already used MCP^[4].
• Current version: revision 2026-07-28 of the specification. Versions are named by the date of the last change that is incompatible with earlier ones^[5].

What does MCP not do on its own?

MCP defines how systems talk, not what is allowed. The specification sets principles of user consent and control, data privacy and tool security, but acknowledges that the protocol cannot enforce them: this falls to whoever implements each application and each server^[1]. Three practical consequences:

• What a connected assistant can do depends on what the server exposes, not on the protocol.
• How confirmations are requested depends on each assistant.
• What reaches the assistant is subject to the terms of that assistant and of the plan contracted.

What changes when you connect MCP to watching?

Without context, an assistant answers with what it already knows or finds at that moment. Connected to a watch, it works on what your team has captured, reviewed and kept, with the source of each signal linked. AI answers; watching expands the context.

Innguma offers connectivity through MCP, currently compatible with ChatGPT, Claude, Microsoft Copilot and Gemini. With the capabilities your organisation enables in its integration, from the assistant you can run the watch cycle: sources, filters, review, publication and queries. What is automated, and what goes through human review first, is decided by your team.