The semantic layer

Why the MCP waits for an index

Before indexing, Kognita has a project, a token, and maybe a git connection, but it does not yet have codebase knowledge. There is nothing reliable for the MCP to retrieve. Once a repository reaches indexed status, the MCP can answer against actual implementation context instead of empty project state.

What indexing creates

Kognita does not treat a repository as a folder of text files. During indexing it builds a searchable representation of the codebase: meaningful chunks, symbols, relationships, repository boundaries, and business-relevant descriptions of what code does. That representation becomes the semantic layer.

• Files become queryable by meaning, not just exact keywords.
• Related code can be retrieved together even when it lives in different folders or repositories.
• The assistant can ask for the narrow part of the system it needs instead of loading broad file dumps.
• The index can be refreshed as repositories change, so answers stay grounded in current code.

Why this lowers token consumption

Without an index, an AI assistant has to guess which files matter, ask for too much context, or rely on whatever the user pasted into the chat. That burns tokens on boilerplate, nearby but irrelevant files, and repeated explanations of the same system. A semantic layer changes the shape of the request.

• The assistant retrieves the most relevant code chunks instead of reading whole directories.
• Repeated project knowledge lives in the index, so every conversation does not need to rebuild it from scratch.
• Fewer irrelevant files enter the prompt, leaving more of the context window for reasoning and the actual task.
• Cross-repo questions can start from likely services and flows instead of scanning every repository manually.

Why this increases useful context

Context window size is not the same as understanding. A bigger prompt full of disconnected files still makes the model reconstruct the system from fragments. Kognita gives the model a better starting point: context that is already organized around meaning, relationships, and behavior.

How to unlock Kognita MCP

• Add at least one repository to the project.
• Run indexing from the Repos tab.
• Wait until at least one repository shows indexed status.
• Return to MCPs and copy the Kognita MCP URL and bearer token into your AI client.