Every agent wants more hands
The Model Context Protocol exists because every agent platform kept needing the same things: tools to call, resources to read, prompts to reuse, and a way for host applications to expose capabilities without bespoke glue for every model.
Tools are actions; resources are context
An MCP tool is something the model can ask the host to execute. An MCP resource is data the host can expose as context, such as files, schemas, or application records. Mixing those up creates sloppy designs.
Resources should be selected and prioritized by the host application. Tools should be permissioned and audited by the host application. The model does not own the boundary.
Why this matters
Without a connector layer, every agent rebuilds the same integrations: GitHub, Slack, Drive, databases, local files, calendars. With a protocol layer, the agent can inherit a capability ecosystem.