Skip to content

Model Context Protocol

MCP standardises exchanges between AI applications and servers providing tools, context data or templates.

Also known asMCP
AI & Automation

What does MCP connect?

The Model Context Protocol, or MCP, is an open protocol for connecting AI applications to available data and functions. A client inside the application communicates with an MCP server. The server may offer executable functions as Tools, context data as Resources and reusable instruction templates as Prompts. The capabilities actually available depend on the implementations on both sides and their compatibility. MCP defines the exchange format; selection, use and access rules must be configured in the specific solution.

In a fictional maintenance system, a server provides approved manuals as readable resources, document search as a tool and a template for follow-up questions. The AI application uses content permitted for the particular user. Such a connection grants neither access to all company data nor automatic authority to change a maintenance work order. The team therefore checks the capabilities actually offered, the supported protocol version and enforcement of access rights. A model may request a tool call through function calling; MCP may provide the shared interface to the corresponding server.

Sources

Related terms

Articles on this topic

Insights · AI & Automation2 minPlan how an AI pilot ends, tooA pilot needs criteria for continuing, narrowing its scope and ending. Functions already developed do not justify the next expansion on their own.Daniel HartmannNov 13, 2026
Insights · AI & Automation2 minAssess language quality and task accuracy separatelyA fluent sentence can name the wrong environment. Assess readability, meaning and the action derived from the text separately in multilingual AI.Philipp ReuterNov 13, 2026
Insights · AI & Automation2 minSeparate confidence from business approvalA high recognition score may correctly reproduce a price. Permission to use that price requires a different basis.Matthias BeckerNov 13, 2026
Insights · AI & Automation2 minDecide separately whether AI reply drafts should be sentA useful reply draft can ease the workload. Automatic sending needs its own justification and a process for review, changes and errors.Matthias BeckerNov 13, 2026
Insights · AI & Automation2 minA model switch also needs a review of behaviourA shared adapter can standardise the connection. Whether a model handles tool failures correctly remains a separate application check.Philipp ReuterNov 13, 2026
Insights · AI & Automation2 minData control extends beyond the AI serverLocal inference can constrain a data flow. A portal using external error logs shows why the rest of the application also needs examination.Nora HenningNov 13, 2026
Insights · AI & Automation2 minCheck whether sources are current before changing the modelTwo invented responses to an old instruction show why valid sources and model quality need separate checks.Matthias BeckerNov 13, 2026
Insights · AI & Automation2 minEvaluate AI answers with read permissions in the test caseWhy a factually correct answer must fail a comparison when it exposes protected information to the person asking.Nora HenningNov 13, 2026

AI & Automation with ASES

AI assistants and automated workflows

Request a quoteMore on AI & Automation