What is an MCP server and how does it work?

Authorize a compatible MCP client to read project context and record supported memory operations. Access respects workspace permissions; an AI answer still needs review.

An MCP server is a component that makes data and functions available to a compatible AI application through the Model Context Protocol. Instead of pasting information into every chat, an assistant can discover declared tools and call one when the user's request requires it.

Protocol, client and server

MCP stands for Model Context Protocol. The standard defines how a host application, client and server negotiate capabilities and exchange requests.

• The host is the application where you work, such as a compatible assistant or editor.
• The client maintains an MCP connection on the host's behalf.
• The server exposes focused capabilities, such as retrieving memory or querying SEO data.

What an MCP server can expose

The specification describes three primary primitives:

• Tools are parameterised functions that retrieve information or perform an action.
• Resources are structured content a client can add to context.
• Prompts are interaction templates a user may select.

Suppose you ask, “Which planned articles are still unwritten, and which existing pages should link to them?” An isolated chat does not know the workspace. Through MCP, the assistant can discover relevant tools, request the necessary records and build its answer from returned evidence.

MCP is not an uncontrolled public API

MCP may use APIs behind the scenes, but it adds a contract designed for AI clients: tool discovery, input descriptions, capability negotiation and structured results. It does not remove authentication, authorisation or consent.

The client should show which tools are available and let the user control their use. A tool that changes records deserves greater scrutiny than one that reads a summary. The protocol enables a call; the product must still enforce identity, permissions and workspace isolation.

What MCP does not do

MCP cannot make incorrect data true, give a model unlimited knowledge or replace a source. If Search Console is not connected, the server cannot invent its metrics. If a page has not been retrieved, it cannot pretend to have read the copy.

Security and authorisation

Before connecting, verify the server identity and requested scope. Use the official authorisation flow, avoid copying tokens into documents and revoke access when it is no longer needed. The assistant should receive only tools and data permitted for that workspace.

Keep a person in control of consequential actions such as creating or updating content. The technical ability to call a tool is not permission to make every possible change.

Frequently asked questions

Does an MCP server contain an AI model?

Not necessarily. It exposes data or functions; the model usually belongs to the client application.

Are MCP and an API the same thing?

No. An MCP server may call APIs, but it presents capabilities to AI clients through a shared standard.

No. It should return the results requested through authorised tools while respecting account and workspace boundaries.

Make context queryable

An MCP server makes a workspace usable by AI without turning it into one endless prompt. The value is not “give everything to the model”; it is declared tools, focused results and an identifiable authorisation boundary.

What BeKnow keeps

Keep interventions, hypotheses, assets, observations and decisions in a searchable workspace. Record what is known and what still needs checking.

A change in performance after an intervention does not prove that the intervention caused it. Cross-platform attribution and a complete analytics dashboard are not available today.

Project memory and data imports do not require an AI model key. A compatible external AI client may have its own costs. BYOK applies only to available functions that actually call an external provider.

Next step

Start with one project, one documented change and the evidence needed to review it. Source connections.

Record your first intervention. How it works.