Introduction to agenticscope
agenticscope is a directory-as-context standard coupled with an integrated Model Context Protocol (MCP) server. It equips AI coding agents with deep, structured awareness of your workspace without needlessly exhausting your context window.
The Context Problem
Currently, developers feed AI agents context via monolithic files (e.g., .cursorrules, CLAUDE.md, GEMINI.md, or AGENTS.md). As projects scale, these files bloat, combining immutable coding standards, database schemas, architectural decisions, and temporary task notes into a single, unwieldy document.
This creates significant friction:
- Token Exhaustion: The agent re-reads irrelevant data (like SQL schemas when modifying CSS) on every inference cycle.
- Priority Dilution: Strict behavioral constraints ("never use
any") are drowned out by sprawling reference materials. - Vendor Fragmentation: Disparate platforms require their own dedicated context files (e.g.,
.claude/,.gemini/), which inevitably drift out of sync.
The agenticscope Solution
agenticscope resolves these challenges through a targeted, indexed approach:
- Typed Fragments: Context is modularized into distinct, typed fragments (
rules,knowledge,specs,personas,memory). - Smart Triggers: Fragments only load when their associated glob paths or keywords match the current task.
- Strict Budgeting: Context is constrained by a strict token budget, dropping lower-priority fragments gracefully.
- Universal Compilation: A single
.scope/directory is compiled natively for multiple vendors instantly (outputting.cursorrules,CLAUDE.md,GEMINI.md, andAGENTS.md). - Live MCP Integration: Advanced clients like Windsurf, Zed, and Cursor can bypass static files entirely. The accompanying MCP server exposes your environment dynamically, letting the agent query active plans, git state, and subagents on demand.
Next, head over to Getting Started to install the CLI and scaffold your first project.