Model Context Protocol: one standard way for AI apps to use outside tools
An agent is as useful as what it can reach. MCP gives every AI application the same way to reach a calendar, a database or a data provider, so a service is connected once and works everywhere.
Model Context Protocol (MCP) is an open-source standard that connects AI applications such as Claude or ChatGPT to outside data and tools through one common interface. Anthropic released it in November 2024. A service with an MCP server can be called by any application that speaks the protocol, with no integration written for each one.
On this page
What MCP connects
Anthropic open-sourced MCP on 25 November 2024 as one protocol in place of the separate integrations each assistant needed for each data source. The official introduction compares it to a USB-C port, one shape of connection with many devices behind it. Claude, ChatGPT, Visual Studio Code and Cursor are among the applications it lists as supporting MCP.
The specification is dated 28 July 2026 in its current version. It defines three roles, which exchange messages in JSON-RPC 2.0, a plain request and response format:
- Host: the AI application a person uses, which starts the connection, such as a chat app or a coding tool.
- Client: a connector inside the host application.
- Server: a service that provides data and actions, such as a company database or a data provider.
A server offers tools (functions the model can run), resources (data for the person or the model) and prompts (templated messages and workflows). When someone asks a question, the model picks a tool and the host calls the server. The specification asks hosts to get the user’s explicit consent before any tool runs.
Why agent builders use MCP servers
For a team building an agent, MCP is how the agent reaches facts it can’t produce itself. A model can reason about a competitor, but it can’t have been signed up to that competitor for a month. With an MCP server for each source, the agent asks in plain words. The same server works in every client that supports the protocol, so nobody writes a connector per app.
How company data reaches an agent over MCP
Obsession collects what companies show only to customers, step by step: their screens, emails and texts, member prices, trial limits, quotes and cancel offers, each dated with a screenshot. Teams and AI agents look up any company through the API or MCP, and a company not yet covered is added on request. Its MCP server gives an agent the same 5 tools as its API.
- lookup: is a company covered, and since when.
- facts: the company’s dated facts, each with its source.
- evidence: the raw email or page behind a fact.
- index: add a company that isn’t covered yet.
- notify: send each new fact to a webhook as it lands.
Connecting takes one command in Claude Code, claude mcp add obsession, with the server address that comes with early access, as the developers page sets out. From then on an agent can ask what any named competitor emailed new customers this week. Each answer comes back with its date and the email itself, for as long as the company is followed.
No. Anthropic started it and released it as open source. The official introduction lists ChatGPT, Visual Studio Code and Cursor among the applications that support it.
An API is a service’s own interface, which a developer writes code against. An MCP server describes the service’s tools in the protocol’s format, so any MCP host can find them and let a model call them without custom code.
The API and MCP for developersThe specification says hosts must get the user’s explicit consent before invoking any tool, and should treat tool descriptions as untrusted unless they come from a trusted server.
Sources
In the order the page cites them. Every page was read on 11 Nov 2026.