AI, audio, and agents
Model Context Protocol
Also called MCP, MCP server
The Model Context Protocol (MCP) is an open standard for connecting AI applications to external tools and data. A service implements an MCP server exposing tools and resources; any compatible client can then discover and call them without a bespoke integration on either side.
The problem it solves
Before a shared protocol, every combination of assistant and service needed its own connector, so integrations grew as a product of both lists. MCP inverts that: a service publishes one server describing its tools, and every compatible client can use it.
A server advertises tools with typed inputs, and optionally resources and prompts. The client discovers what is available at connection time, which is why a new capability can appear in an assistant without the assistant shipping an update.
What it means for reading tools
Retrieval is a natural MCP tool. An assistant asked about a URL can call a reader instead of guessing from its training data, and receives clean article text with the source URL attached — grounded material rather than recollection.
In smry
smry runs a public read-only MCP endpoint at api.smry.ai/mcp exposing a get_article tool, so an assistant can fetch a readable version of a public URL without an API key.
Common questions
Do I need an API key to use smry's MCP server?
No. The public endpoint is read-only and exposes article retrieval without a key.
Is MCP tied to one AI vendor?
No. It is an open specification with implementations across multiple clients and a large ecosystem of independent servers.
How is an MCP tool different from a plugin?
A plugin is built for one host application. An MCP server is described once in a shared format and works with any compatible client.