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

What is an MCP client? Its role between a host and a server

Understand MCP clients through an editor connecting to two data services, including host roles, discovery, version compatibility, permissions, and end-to-end checks.

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

Understand the concept

An MCP client is a protocol component in an AI host that communicates with a server, discovers capabilities, and exchanges requests and results. Hosts can manage clients for different services. A client is neither the chat user nor the model. Connectivity alone does not establish supported tools, permissions, or correct data.

01

Hosts manage clients rather than being identical to them

An editor can manage separate clients for issue and file services behind one assistant interface. Connections and permissions can differ. One working service does not establish authentication or equivalent capability for another.

Official architecture defines clients as components obtaining server context. Servers can be local or network-accessed. Client location differs from user identity, and remote connections do not inherently gain greater privilege.

02

Discovery and version compatibility govern interactions

Clients need supported versions and capabilities before calling tools or reading resources. Older implementations may lack newer features. Check protocol compatibility before blaming failures on model tool-use ability.

Listing a tool is followed by checking arguments and return formats. Hosts may expose capabilities to models, while runtimes process requests. Protocol exchange differs from model decisions, and clients do not guarantee business conclusions.

03

Check authorization and results end to end

Validate an allowed read-only request end to end and confirm out-of-scope resources are blocked. This is a proposed check without real account connections. Preserve request-result associations to avoid assigning stale results to new tasks.

Fetched text retains its source authority rather than becoming user authorization. Hosts decide context use and servers enforce access. Successful communication does not establish truth; trace final claims and verify target state for writes.

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 ↗