Context Master for VS Code

Deterministic AST-based context for coding agents.

Context Master runs an MCP server inside VS Code and exposes opinionated and streamlined tools that let agents navigate along the edges of the AST tree using data obtained from the VSCode API. This prevents hallucinations or ambiguous results from regular grepping through files.

Why Context Master?

Modern AI assistants are powerful, but they often lack the deep understanding of your project's architecture needed to generate high-quality, context-aware code. Context Master solves this by providing the AI with the same deep-level insights a human developer would have. This gives you:

• Agents that reason over symbol relationships instead of scanning random files.
• Lower hallucination risk in large codebases.
• Better edit accuracy for refactors, call-site updates, and impact analysis.
• Deterministic context retrieval based on VS Code language services.
• More efficient token management through symbol-based source code retrieval instead of file-based retrieval.
• Better planning results through impact analysis and deterministic discovery process.

How It Works

Context Master starts a local MCP server that your AI assistant can connect to. This server exposes a set of powerful tools that allow the AI to query your codebase using the full power of VS Code's language intelligence. The AI will follow along traceable edges of the AST tree and thus get a complete and accurate picture of the codebase.

Biggest Capabilities

• AST/graph-backed context engine: structural repository understanding, not vector retrieval or regex guessing.
• High-level MCP tools: we don't blindly expose the VSCode API to your agent. Instead, we expose a set of opinionated tools that are tailored for a proper workflow.
• Token efficiency: agents can request exactly what they need.
• Free mode available: evaluate before buying.

For a more in-depth explanation of how Context Master works, please refer to our website's how it works page.

MCP Tools Exposed

• get-repository-overview: Returns the most architecturally significant symbols in the codebase, ranked by multiple centrality metrics. Helps to build a mental map of the codebase and identify the key files and symbols relevant to the user's request.
• get-file-tree: Returns a list of files in the workspace, their top-level symbols and what other symbols these files import. Helps agents understand the file structure and the relationships between files. Get's ineffective in large codebases.
• get-symbol-context: This is the heart of Context Master. It returns the full source code of a symbol, its relationships to other symbols, information about its children and ancestors and thus all the context, the agent might need regarding one specific Symbol. A better, more targeted alternative to reading the whole file.
• read-file: Returns the full content of a file plus meta information about its top.level symbols as well as imported symbols.
• read-many-files: The same as read-file, but for multiple files.

Quick Start

1. Install from the VS Code Marketplace.
2. Open the Context Master dashboard and check the Ports tab for the mapped MCP server port of the currently open workspace.
3. Connect your MCP-capable assistant to http://localhost:<mapped-port> by adding the config manually or just asking your coding agent to configure it for you. The default port is 13692.

Example Cursor MCP config:

{
  "mcpServers": {
    "context-master": {
      "url": "http://localhost:13692"
    }
  }
}

AGENTS.md Helper

The extension includes a built-in AGENTS.md installer:

• Open Dashboard (click "CM"-badge in lower left corner) -> Install AGENTS.md Rules
• or run command: Context Master: Install AGENTS.md Rules into Current Workspace

Behavior:

• If AGENTS.md does not exist in the active workspace folder, it is created.
• If it exists, Context Master rules are appended to the end.
• Existing user content is preserved (no overwrite).

Free vs Licensed

• Free mode: up to 25 MCP tool calls per day
• Licensed mode: unlimited tool calls

Key Settings

Command Palette Commands

• Context Master: Open Dashboard
• Context Master: Install AGENTS.md Rules
• Context Master: Restart MCP Server
• Context Master: Show extension logs
• Context Master: Generate repository overview for a specific workspace folder
• Context Master: Trigger indexing of a workspace folder
• Context Master: Check your license status
• Context Master: Reset your license status to free mode
• Context Master: Open Privacy Policy

Privacy & Telemetry

Context Master respects VS Code's global telemetry setting. If telemetry is disabled in VS Code, the extension does not send any telemetry or analytics data. Your source code or any other sensible data never leaves your machine, not even graph state or topology are logged anywhere but locally on your machine.

Privacy policy: https://context-master.dev/privacy

Feedback

Report issues or requests at context-master-issues.

Troubleshooting

Here are some answers to frequently asked questions:

Q: How do I configure my AI Assistant (e.g., Cursor) to use Context Master?

A: Open Context Master dashboard -> Ports and copy the mapped port for your current workspace. Add that port to your assistant MCP configuration:

{
  "mcpServers": {
    "context-master": {
      "url": "http://localhost:<mapped-port>"
    }
  }
}

Each VS Code window runs its own MCP server instance. Ports are stable per workspace and stored in a machine-local Port Map.

Q: Why do I need to configure different ports per VSCode Workspace?

A: Each workspace uses its own MCP server instance to keep workspace context isolated. This avoids forcing the LLM to choose the correct workspace on every interaction, which would otherwise require adding an extra workspace parameter to each tool call. Keeping ports separate reduces decision overhead and keeps calls simpler and more deterministic. We may revisit this approach in the future if it becomes clear that LLMs can reliably handle the added multi-workspace complexity.

Q: The extension seems stuck on indexing. What should I do?

A: Initial indexing can take some time for large projects. You can see the progress in the notification area. If it seems stuck, try the following:

1. Open the logs using the Context Master: Show extension logs command to check for any errors.
2. Ensure you have the necessary language extensions installed for your project (e.g., the official Python or TypeScript extensions).
3. Try restarting VS Code.

Q: The mapped port is already in use. What should I do?

A: Context Master prompts you with three options: Increase to the next suggested port, Open Port Map to edit assignments, or Retry after you manually free the port. The selected mapping is persisted in the Port Map.

Q: Does Context Master support multi-root workspaces?

A: Yes, there is preliminary support for multi-root workspaces. However, tool calls are currently scoped to a single workspace folder at a time. This means the AI cannot yet see a structurally linked overview of all your open folders at once. If you'd like to see more advanced multi-root support, please let us know by opening an issue or starting a discussion on our GitHub repository!

Q: Why isn't my AI assistant using the tools from Context Master? A: First, verify your MCP server configuration in your AI assistant's settings. Second, check the VS Code status bar to ensure the Context Master server is running. If you see an error, check the extension logs. Finally, consult your AI assistant's documentation to see how it discovers and uses MCP tools.

Q: How can I inspect what the extension is doing?

A: Use Context Master: Show extension logs and, if needed, increase verbosity via contextMaster.logLevel.

Q: I found a bug or have a feature request. Where should I report it?

A: We'd love to hear from you! Please head over to our GitHub repository to open an issue or start a discussion.