hotai.news
← Back to AI WikiAI Wiki / Core concepts

What is an MCP resource? Readable context versus executable tools

Understand resource URIs, types, and templates through a project file, distinguishing listing, reading, context inclusion, and authority.

HotAI.news · Sources checked with CodexPublished on 2026-10-20Sources checked on 2026-10-20
Our take

Understand the concept

An MCP resource is contextual data exposed by a server, such as file content or database records, typically identified by a URI. It focuses on reading context rather than performing actions. Listing does not establish reading or automatic full model inclusion; permissions, content types, and context use depend on implementations.

01

Resource identifiers locate data rather than execute actions

A project description can be read by URI and then selected by the host for answering. URIs can use file-like or service-specific forms; not every resource opens in a browser, and a path name does not establish access.

The specification supports text and binary resources with media information. Handle actual types instead of interpreting encoded binary as prose. A link or listing still requires successful content retrieval before claiming review.

02

Resource templates and tool arguments can look similar

Templates locate content by a pattern, such as project ID, without guaranteeing every ID exists. Tools can search or modify data; even textual results retain action semantics defined by the tool.

For current data, check update and notification support. Stable URIs do not imply immutable content, and cached copies can be stale. Versions and read times distinguish server updates from unrefreshed host content.

03

Keep source and permission boundaries after reading

Instructions inside resources retain their source status rather than becoming new user commands. Reading one public file does not authorize private neighbors. Programs enforce access; prompts only assist content separation.

Start with known short content, check type and values, and verify unauthorized reads are denied. This is a validation design without a real file service. Retain traceable locations; MCP delivery alone does not establish source reliability.

Checked sources

These sources support the concepts and methods described here. Examples and test scope are labeled separately; check the version of the tools you use.

Back to AI Wiki ↗