The Model Context Protocol (MCP) is an open protocol, introduced by Anthropic in late 2024 and since adopted by many AI vendors and developer tools, that standardizes how AI applications connect to outside systems. A company publishes an MCP server describing the tools and data it offers, and any compatible AI client can use them. For SaaS companies, an MCP server is a way to make the product usable directly from assistants and agents.
How MCP works
MCP defines a client-server relationship. The AI application is the client. Each integration is a server that exposes capabilities in a standard format: tools the model can call, resources it can read and prompts it can use. Because the format is shared, a SaaS company can build one server instead of a separate plugin for every assistant.
- Tools: actions such as 'create invoice' or 'search tickets'
- Resources: data the model can read, such as documents or records
- Prompts: reusable templates the server suggests
Why it matters beyond engineering
As assistants become a place where work happens, being reachable from them becomes a distribution question. If a user asks an assistant to 'create a project from these notes', the assistant can only use products it is connected to. MCP servers, and the directories that list them, are an emerging discovery surface. How AI products choose which connectors to suggest is not well documented, so treat this as a channel to watch and test.
MCP servers also raise security questions: they grant AI systems access to real data and actions, and prompt injection risks apply.
Example
A product usable from any assistant
Marrowgate, a fictional CRM, ships an MCP server with tools to look up contacts, log calls and create deals. Customers connect it to the assistants they already use. Sales reps start logging calls by asking their assistant, and the integration becomes a selling point listed on Marrowgate's integrations page.
Common mistakes
Exposing too much
Scope tools tightly, require authentication and confirm destructive actions. An agent with broad write access is a real risk.
Shipping without docs
An MCP server still needs a public page explaining what it does, how to connect it and what permissions it needs.
Questions
No. It is an open specification, and many AI clients and development tools from different vendors support it.
They address different needs. llms.txt is a proposed file for describing site content to models. MCP lets models call tools and fetch data through a server.
