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

What is an MCP server? Exposing tools and contextual capabilities

Understand MCP servers through project data, including local and remote deployment, capability declarations, access to underlying data, and result handling.

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

Understand the concept

An MCP server is a program exposing capabilities such as tools, resources, and prompts through the protocol. It can be local or remote and need not occupy a separate machine. Running one does not retrain models. Interface capabilities and underlying account scope must be explicit; installation does not establish successful queries.

01

Servers wrap capabilities rather than replace data sources

A project server can wrap an issue API as a tool and expose documentation as a resource. The underlying system still governs account access. MCP does not automatically narrow an overprivileged account to the current user; implementations need identity mapping and scope controls.

Names and descriptions should reflect real actions rather than conceal querying and deletion behind one vague operation. Clients discover capabilities and models interpret arguments. Clear documentation helps selection but does not eliminate program validation.

02

Local and remote deployments have different requirements

Official architecture distinguishes stdio from Streamable HTTP. The former commonly serves host-launched local processes; the latter supports remote services. This describes architecture rather than installation commands, which must match the implementation and version.

Local services can still access networks and files and are not inherently trusted. Remote ones need authentication and data-scope checks. Inspect provenance and dependencies rather than granting entire directories or cloud accounts merely to connect.

03

Validate real tool and resource paths

Discover capabilities, read known material, and run an allowed read-only query with field checks. No real server was started here. A running process does not establish an available external API; preserve errors instead of treating empty text as success.

Write tools need idempotency and cancellation, while returned instructions do not expand authorization. Recheck critical paths after upgrades to avoid stale host assumptions. This page explains responsibilities without endorsing arbitrary services’ security or compatibility.

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 ↗