Giving AI Agents the Context They Need

AI coding assistants like Claude Code, Cursor, and GitHub Copilot are remarkably capable. They can write functions, debug code, refactor entire files. But they all share the same limitation: they don’t know what you’ve been doing.

Every conversation starts from zero.

The Context Gap

When you ask an AI assistant to help with a bug, it doesn’t know:

• That you spent the last two hours in the authentication module
• That you switched between three different files
• That you had a Slack conversation about this exact issue
• That you looked at the production logs twenty minutes ago

You have to explain all of this. Every time. And the explanation is never as good as actually having been there.

This is the AI context problem. Your AI assistant is smart but amnesiac. It has no memory of your work history, no understanding of your current project state, no awareness of what you’ve been trying to accomplish.

Why Context Matters for AI Productivity

Studies show that developers spend up to 30% of their time re-explaining context to AI tools. That’s not just inefficient—it’s frustrating. The promise of AI assistants is that they help you work faster, but if you’re constantly explaining background, you’re working slower.

The best human colleagues understand your project context. They know what you’ve been working on, what problems you’ve hit, what decisions you’ve made. That context makes their help actually helpful.

AI assistants need the same thing.

What Stubble Enables

Stubble maintains a timeline of your work—what apps you used, what files you edited, what meetings you attended. This context can be shared with AI agents via MCP (Model Context Protocol) to give them the background they need.

Instead of starting from scratch, an AI assistant can understand:

• “You’ve been working on the auth module for 2 hours”
• “You switched between UserService.ts and AuthController.ts repeatedly”
• “Your last meeting discussed the OAuth migration”
• “You searched for ‘token refresh’ in the logs”

This context transforms generic AI into a genuine assistant that understands your specific situation.

How It Works

Stubble exposes your work context through a local MCP server. AI tools that support MCP (like Claude Code) can query:

• Recent activity by time range
• Files and apps used in the current session
• Project associations based on folder patterns
• Meeting context from calendar integration
• OCR digests showing what you were looking at

The AI agent asks for context, Stubble provides it, and suddenly the assistant knows what you know.

Example: Claude Code Integration

When you use Claude Code with Stubble connected, your prompts automatically include relevant context:

You: "Help me fix this authentication bug"

Claude (with Stubble context): "I see you've been working on UserService.ts
and AuthController.ts for the past 2 hours, and you had a meeting about
OAuth migration this morning. Based on the files you've been editing,
it looks like the issue might be in the token refresh logic..."

No manual explanation needed. The AI already knows your context.

Privacy & Security

Stubble’s MCP server includes multiple security layers:

1. Opt-in — MCP access is disabled by default. You explicitly enable it in the Connect tab.
2. API key auth — Every request requires your personal API key
3. Data sanitization — All text is sanitized to remove secrets (API keys, passwords, tokens) before being shared
4. Rate limiting — 60 requests per minute prevents abuse
5. Audit logging — Every access is logged so you can see what was shared

The AI tools receive sanitized context about your work patterns—not raw screenshots or sensitive credentials.

What Gets Shared (and What Doesn’t)

Shared with AI tools:

• App names and window titles (sanitized)
• File paths (basename only)
• Task summaries generated by AI
• Project names and time allocations
• Meeting titles and attendees

Never shared:

• Raw screenshots
• Full file contents
• Passwords, API keys, tokens
• Email addresses
• Meeting transcripts
• Granola meeting notes

Connecting Your AI Tools

Stubble supports multiple AI tools through MCP:

Claude Code:

claude mcp add-json stubble '{"type":"stdio","command":"/Applications/Stubble.app/Contents/MacOS/stubble-mcp","env":{"STUBBLE_MCP_KEY":"your-key"}}' --scope user

Cursor: Add to ~/.cursor/mcp.json

Windsurf: Add to ~/.codeium/windsurf/mcp_config.json

Claude Desktop: Install the Stubble.mcpb extension

Setup instructions are in the Stubble app under the Connect tab.

The Future of AI Assistants

Right now, AI assistants are powerful but amnesiac. Every conversation is a fresh start. Every explanation is from scratch.

Context-aware AI changes this. When your assistant knows what you’ve been working on, it can:

• Suggest relevant code based on recent changes
• Understand the project structure you’ve been navigating
• Reference conversations and decisions from earlier
• Build on work instead of starting over

This is where developer tools are heading. Not just smarter models—smarter context.

Download Stubble for Mac and give your AI tools the context they need.