Kognyt exposes Salesforce intelligence through the extension and through the bundled MCP server. These are related, but they are not the same thing.
1. Extension-native use in VS Marketplace-compatible IDEs
The Kognyt extension itself can run in VS Code, Cursor, and other IDEs that support VS Code extensions.
That gives you the extension-native experience:
- Kognyt sidebar and org management
- sign-in and connected-org flows
- status bar state
- retrieve, deploy, validation, and drift commands
This is the direct extension path. It is separate from MCP.
2. GitHub Copilot in VS Code
Kognyt registers Salesforce capabilities as VS Code Language Model Tools. In supported Copilot workflows, those tools can be referenced in prompts and invoked by the model.
3. MCP-based use through the bundled Kognyt MCP server
Kognyt also ships with a built-in MCP server. The extension can configure MCP access for detected hosts and AI tools that consume MCP, including:
- Cursor
- Claude Code
- Codex
- Cline
- Kilo Code
This MCP path is what lets those tools use Kognyt's Salesforce context through the extension's bundled server.
Important distinction
- VS Code and Cursor can run the Kognyt extension directly
- Some tools also consume Kognyt through MCP configuration
So a tool or environment can be relevant in more than one way. Cursor is the clearest example: it can run the extension directly, and it also has an MCP configuration path.
Kilo Code and Cline are detected as AI-tool integrations that the extension can wire up to the bundled MCP server.
Claude Code and Codex are handled as separate CLI-based MCP integration paths.
Use Kognyt: Setup AI Provider when you need to run the setup flow manually.
Auto-configuration behavior
On activation, Kognyt detects supported providers and offers to configure MCP for them. It writes the Kognyt MCP entry while keeping a stable launcher path across extension upgrades.
Data handling
To ground answers in what you are working on, Kognyt sends the active file's path, your selection, the Salesforce entities it detects, and the file's content to the Kognyt backend. Connecting an org also syncs and stores its metadata and source so the tools can reason about it. This data is encrypted in transit, stored under least-privilege access, and never used to train shared models. See the Privacy and Security pages for details.