@sashabogi/argus-mcp 2.0.1 → 2.0.3

package/README.md

@@ -1,570 +1,270 @@
 # Argus
 
-![Argus Banner](assets/argus-banner.jpg)
+**Understand Any Codebase, Regardless of Size**
 
-**Codebase Intelligence Beyond Context Limits**
+Argus solves a fundamental problem: **your codebase is too big to fit in an LLM's context window**. You can't just paste 50,000 lines of code into Claude and ask "how does authentication work?" - it won't fit, and even if it did, the AI would get lost in the noise.
 
-Argus is an AI-powered codebase analysis tool that understands your entire project, regardless of size. Using a hybrid approach of zero-cost structural queries and AI-powered reasoning, Argus provides intelligent answers about code architecture, patterns, and relationships that would be impossible with traditional context-limited approaches.
+Argus creates intelligent **snapshots** of your codebase and provides tools that let Claude explore them efficiently. Instead of trying to load everything at once, Claude can search, navigate, and analyze your code piece by piece - just like a human developer would.
 
-## Key Features
-
-| Feature | Description |
-|---------|-------------|
-| **Zero-Cost Dependency Queries** | Find importers, symbols, and file dependencies instantly with pre-computed metadata |
-| **Progressive Disclosure Search** | 3-tier search architecture: `find_files` -> `search_codebase` -> `get_context` |
-| **Enhanced Snapshots** | Import graphs, export indexes, and symbol lookups built into every snapshot |
-| **Worker Service** | Persistent LRU caching for repeated queries (port 37778) |
-| **Web UI** | Interactive D3.js dependency graph visualization |
-| **Multi-Provider AI** | ZAI, Anthropic, OpenAI, DeepSeek, and Ollama support |
-| **Claude Code Integration** | Native MCP server with global instructions for all agents |
-
-## Quick Start
+![Codebase Explorer](docs/images/codebase-explorer.png)
 
-```bash
-# 1. Install globally via npm
-npm install -g @sashabogi/argus-mcp
+## How It Works
 
-# 2. Interactive setup (configures API keys and preferences)
-argus init
+### 1. Create a Snapshot
 
-# 3. Add to Claude Code (installs MCP + global instructions)
-argus mcp install
+A snapshot is a single text file containing your entire codebase, organized for efficient searching:
 
-# 4. Set up any project
-argus setup .
+```bash
+argus snapshot . -o .argus/snapshot.txt
 ```
 
-That's it! Argus is now integrated into Claude Code and will automatically be used for codebase exploration.
+This creates a ~300KB file from a typical project that includes:
+- All your source files with line numbers
+- **Import graph** - which files depend on which
+- **Export index** - where every function/class is defined
 
-## Installation
+### 2. Claude Searches Before Reading
 
-### npm (Recommended)
+When you ask Claude about your code, it doesn't read the whole snapshot. Instead, it uses **progressive disclosure**:
 
-```bash
-npm install -g @sashabogi/argus-mcp
 ```
+You: "How does the authentication flow work?"
 
-### Manual Installation
-
-```bash
-git clone https://github.com/sashabogi/argus.git
-cd argus
-npm install
-npm run build
-npm link
+Claude's process:
+1. search_codebase("auth")     → Finds 12 matches in 4 files (FREE - no AI)
+2. get_context("auth.ts", 42)  → Gets 20 lines around the match (FREE)
+3. find_importers("auth.ts")   → Sees what depends on it (FREE)
+4. analyze_codebase("...")     → Deep analysis only if needed (~500 tokens)
 ```
 
-### Update
+**90% of questions are answered with FREE tools.** The AI-powered analysis is only used for complex architectural questions.
 
-```bash
-argus update
-```
-
----
+### 3. Automatic Updates
 
-## Features
+You don't need to manually refresh snapshots. Argus integrates with Claude Code's lifecycle:
 
-### 1. Progressive Disclosure MCP Tools
+- **Session start**: Checks if snapshot exists and is fresh
+- **After file edits**: Flags snapshot as stale
+- **On your command**: `argus snapshot .` to refresh
 
-Argus provides 6 MCP tools with a tiered cost structure, optimized for token efficiency:
-
-| Tool | Cost | Use Case |
-|------|------|----------|
-| `search_codebase` | **FREE** | Fast regex search across the entire snapshot |
-| `find_files` | **FREE** | Glob pattern matching for file discovery |
-| `get_context` | **FREE** | Extract N lines around a specific location |
-| `find_importers` | **FREE** | Find all files that import a given module |
-| `find_symbol` | **FREE** | Locate where a symbol is exported from |
-| `get_file_deps` | **FREE** | Get all imports of a specific file |
-| `analyze_codebase` | ~500 tokens | AI-powered deep analysis for complex questions |
-| `semantic_search` | ~100 tokens | Natural language code search using FTS5 |
-| `create_snapshot` | ~100 tokens | Create or refresh a codebase snapshot |
+---
 
-**Workflow Example:**
+## Quick Start
 
 ```bash
-# Step 1: Find files (FREE)
-find_files(".argus/snapshot.txt", "**/*auth*")
-
-# Step 2: Search for patterns (FREE)
-search_codebase(".argus/snapshot.txt", "handleAuth")
-
-# Step 3: Get context around a match (FREE)
-get_context(".argus/snapshot.txt", "src/auth.ts", 42, 10, 20)
-
-# Step 4: Only if needed - AI analysis (~500 tokens)
-analyze_codebase(".argus/snapshot.txt", "How does the authentication flow work?")
-```
+# Install
+npm install -g @sashabogi/argus-mcp
 
-### 2. Enhanced Snapshots
+# Setup (configures AI provider for deep analysis)
+argus init
 
-Every snapshot includes structural metadata that enables zero-cost dependency queries:
+# Add to Claude Code
+argus mcp install
 
-```bash
-# Create an enhanced snapshot (default)
+# Create your first snapshot
+cd your-project
 argus snapshot . -o .argus/snapshot.txt
-
-# Create a basic snapshot (faster, smaller, no metadata)
-argus snapshot . -o .argus/snapshot.txt --basic
-```
-
-**Metadata Included:**
-
-- **Import Graph** - Which files import which other files
-- **Export Index** - Symbol -> files that export it
-- **Who Imports Whom** - Reverse dependency graph for impact analysis
-- **Function Signatures** - With line numbers for precise navigation
-
-**Example Zero-Cost Queries:**
-
-```typescript
-// Find where useAuth is exported from
-find_symbol(".argus/snapshot.txt", "useAuth")
-// -> { file: "contexts/auth-context.tsx", line: 42 }
-
-// Find all files that depend on auth-context
-find_importers(".argus/snapshot.txt", "auth-context")
-// -> ["app.tsx", "dashboard.tsx", "settings.tsx"]
-
-// Get dependencies of a specific file
-get_file_deps(".argus/snapshot.txt", "src/app.tsx")
-// -> ["./auth", "./theme", "@/components/ui"]
 ```
 
-### 3. Worker Service
+That's it. Now when you ask Claude about your code, it will automatically use Argus.
 
-For optimal performance, Argus includes a persistent worker service that caches parsed snapshots in memory:
+---
 
-```bash
-# Start the worker service
-argus worker start
+## The Tools
 
-# Check worker status
-argus worker status
+Argus provides 6 MCP tools, organized by cost:
 
-# Stop the worker
-argus worker stop
-```
+### Zero-Cost Tools (No AI, Instant Results)
 
-**Worker Features:**
+| Tool | What It Does |
+|------|--------------|
+| `search_codebase` | Regex search across all files. Returns matches with line numbers. |
+| `find_files` | Glob pattern matching. Find files by name pattern. |
+| `get_context` | Extract lines around a specific location. Like `grep -C`. |
+| `find_importers` | What files import this module? Uses pre-computed import graph. |
+| `find_symbol` | Where is this function/class defined? Uses export index. |
+| `get_file_deps` | What does this file import? |
 
-- **LRU Cache** - Keeps up to 5 parsed snapshots in memory
-- **File Watching** - Detects changes and invalidates cache automatically
-- **REST API** - Available on port 37778
+### AI-Powered Tools (Uses Tokens)
 
-**API Endpoints:**
+| Tool | Cost | What It Does |
+|------|------|--------------|
+| `analyze_codebase` | ~500 tokens | Deep analysis using recursive reasoning. For complex questions. |
 
-| Endpoint | Method | Description |
-|----------|--------|-------------|
-| `/health` | GET | Health check and cache stats |
-| `/snapshot/load` | POST | Load a snapshot into cache |
-| `/search` | POST | Search using cached snapshot |
-| `/context` | POST | Get context from cached snapshot |
-| `/watch` | POST | Start watching a project |
-| `/watch` | DELETE | Stop watching a project |
+**The key insight**: Most developer questions ("where is X defined?", "what calls Y?", "show me the auth code") can be answered with zero-cost tools. You only need AI for questions requiring reasoning across multiple files.
 
-### 4. Web UI - Codebase Explorer
+---
 
-Argus includes a visual codebase explorer for understanding your project structure:
+## Snapshots: Basic vs Enhanced
 
 ```bash
-# Open the web UI
-argus ui
+# Enhanced (default) - includes import graph and export index
+argus snapshot .
 
-# Specify a custom port
-argus ui --port 4000
+# Basic - just the code, faster to create
+argus snapshot . --basic
 ```
 
-![Codebase Explorer](docs/images/codebase-explorer.png)
-
-**Web UI Features:**
+**Enhanced snapshots** enable the zero-cost dependency tools (`find_importers`, `find_symbol`, `get_file_deps`). They take slightly longer to create but make queries much more efficient.
 
-- **Dependency Graph** - D3.js force-directed visualization of file relationships
-- **File Tree Explorer** - Navigate your codebase with line counts
-- **Full-Text Search** - Search across all files with highlighting
-- **Code Viewer** - View file contents with syntax highlighting
-- **Stats Dashboard** - Overview of files, imports, and exports
-
-### 5. Claude Lifecycle Hooks
-
-Argus can automatically update snapshots using Claude Code's lifecycle hooks:
-
-**Session Start Hook:**
-- Checks for `.argus/snapshot.txt` on session start
-- Warns if snapshot is stale (>1 hour old)
-
-**Post-Tool Hook:**
-- Fires after Write/Edit operations
-- Notifies worker service of file changes for cache invalidation
+**Basic snapshots** only support regex search. Use these for very large codebases where you don't need dependency analysis.
 
 ---
 
-## CLI Commands
+## Web UI: Codebase Explorer
 
-| Command | Description |
-|---------|-------------|
-| `argus init` | Interactive setup wizard for API keys and preferences |
-| `argus setup [path]` | One-command project setup (snapshot + CLAUDE.md + .gitignore) |
-| `argus snapshot <path>` | Create a codebase snapshot |
-| `argus analyze <path> <query>` | Analyze codebase with AI |
-| `argus query <snapshot> <query>` | Query an existing snapshot |
-| `argus search <snapshot> <pattern>` | Fast grep search (no AI) |
-| `argus status [path]` | Check if snapshot is up to date |
-| `argus mcp install` | Install MCP server for Claude Code |
-| `argus mcp uninstall` | Remove from Claude Code |
-| `argus context generate <path>` | Generate architecture summary |
-| `argus context inject <path>` | Add architecture section to CLAUDE.md |
-| `argus config [key] [value]` | View or modify configuration |
-| `argus ui` | Open the web UI for visualization |
-| `argus update` | Update Argus to the latest version |
-
-### Command Options
-
-**`argus snapshot`**
-```bash
-argus snapshot ./my-project -o .argus/snapshot.txt
-  --extensions, -e    File extensions to include (comma-separated)
-  --exclude           Patterns to exclude (comma-separated)
-  --output, -o        Output file path
-  --basic             Skip structural metadata (faster, smaller)
-```
-
-**`argus analyze`**
-```bash
-argus analyze ./my-project "What are the main modules?"
-  --provider, -p      Override default provider
-  --max-turns, -t     Maximum reasoning turns (default: 15)
-  --verbose, -v       Show detailed execution logs
-```
-
-**`argus search`**
-```bash
-argus search .argus/snapshot.txt "authentication"
-  --ignore-case, -i   Case-insensitive search
-  --max-results, -n   Maximum results (default: 50)
-```
+Visualize your codebase structure:
 
-**`argus ui`**
 ```bash
 argus ui
-  --port, -p          Port to serve on (default: 3333)
-  --no-open           Do not open browser automatically
 ```
 
----
-
-## MCP Server Setup
+Opens a browser with:
+- **Dependency graph** - Interactive D3.js visualization of imports
+- **File explorer** - Browse files with line counts
+- **Search** - Full-text search with highlighting
+- **Code viewer** - Read any file from the snapshot
 
-### Automatic Installation
-
-```bash
-argus mcp install
-```
+Drag and drop a snapshot file, or paste its contents directly.
 
-This command:
-1. Creates a wrapper script at `~/.argus/argus-mcp-wrapper`
-2. Registers the MCP server with Claude Code
-3. Injects Argus instructions into `~/.claude/CLAUDE.md` (global for all projects)
+---
 
-### Manual Installation
+## Keeping Snapshots Fresh
 
-If automatic installation fails, add manually:
+### Manual Refresh
 
 ```bash
-claude mcp add argus -s user -- ~/.argus/argus-mcp-wrapper
-```
+# Check if snapshot is stale
+argus status .
 
-### MCP Tool Reference
-
-Once installed, Claude Code has access to these tools:
-
-**`search_codebase`** - Zero-cost regex search
-```typescript
-search_codebase({
-  path: ".argus/snapshot.txt",
-  pattern: "handleAuth",
-  caseInsensitive: true,   // default: true
-  maxResults: 50,          // default: 50, max: 200
-  offset: 0,               // for pagination
-  contextChars: 80         // truncate context
-})
+# Refresh
+argus snapshot . -o .argus/snapshot.txt
 ```
 
-**`find_files`** - Zero-cost glob matching
-```typescript
-find_files({
-  path: ".argus/snapshot.txt",
-  pattern: "**/*.test.ts",
-  caseInsensitive: true,   // default: true
-  limit: 100               // default: 100, max: 500
-})
-```
+### Automatic (Claude Code Integration)
 
-**`get_context`** - Zero-cost context extraction
-```typescript
-get_context({
-  path: ".argus/snapshot.txt",
-  file: "src/auth.ts",
-  line: 42,
-  before: 10,              // lines before
-  after: 20                // lines after
-})
-```
+When you run `argus mcp install`, Argus adds lifecycle hooks:
 
-**`find_importers`** - Zero-cost dependency query
-```typescript
-find_importers({
-  path: ".argus/snapshot.txt",
-  target: "src/auth.ts"
-})
-```
+1. **Session start**: If no snapshot exists or it's >24h old, prompts you to create one
+2. **After edits**: The snapshot is flagged as potentially stale
 
-**`find_symbol`** - Zero-cost symbol lookup
-```typescript
-find_symbol({
-  path: ".argus/snapshot.txt",
-  symbol: "useAuth"
-})
-```
+You can also add to your project's `CLAUDE.md`:
 
-**`get_file_deps`** - Zero-cost import list
-```typescript
-get_file_deps({
-  path: ".argus/snapshot.txt",
-  file: "src/app.tsx"
-})
-```
+```markdown
+## Codebase Intelligence
 
-**`analyze_codebase`** - AI-powered analysis (~500 tokens)
-```typescript
-analyze_codebase({
-  path: ".argus/snapshot.txt",
-  query: "How does authentication work?",
-  maxTurns: 15             // default: 15
-})
-```
+This project uses Argus. The snapshot is at `.argus/snapshot.txt`.
 
-**`semantic_search`** - Natural language search
-```typescript
-semantic_search({
-  path: "/path/to/project",
-  query: "authentication middleware",
-  limit: 20                // default: 20
-})
+Before answering questions about the code:
+1. Use `search_codebase` to find relevant files
+2. Use `get_context` to read specific sections
+3. Only use `analyze_codebase` for complex architectural questions
 ```
 
 ---
 
 ## Configuration
 
-Argus stores configuration in `~/.argus/config.json`. Run `argus init` for interactive setup.
+Argus stores config in `~/.argus/config.json`:
 
 ```json
 {
-  "provider": "anthropic",
+  "provider": "ollama",
   "providers": {
-    "anthropic": {
-      "apiKey": "sk-ant-...",
-      "model": "claude-sonnet-4-20250514"
-    },
-    "zai": {
-      "apiKey": "your-api-key",
-      "model": "glm-4.7",
-      "endpoint": "https://api.z.ai/api/coding/paas/v4"
-    },
-    "openai": {
-      "apiKey": "sk-...",
-      "model": "gpt-4o"
-    },
-    "deepseek": {
-      "apiKey": "your-api-key",
-      "model": "deepseek-chat"
-    },
     "ollama": {
-      "baseUrl": "http://localhost:11434",
       "model": "qwen2.5-coder:7b"
     }
-  },
-  "defaults": {
-    "maxTurns": 15,
-    "turnTimeoutMs": 60000,
-    "snapshotExtensions": ["ts", "tsx", "js", "jsx", "rs", "py", "go", "java"],
-    "excludePatterns": ["node_modules", ".git", "dist", "build"]
   }
 }
 ```
 
-### Supported Providers
-
-| Provider | Models | Best For |
-|----------|--------|----------|
-| **ZAI** | GLM-4.7, GLM-4.6 | Best value, excellent coding |
-| **Anthropic** | Claude Sonnet/Opus | Highest quality reasoning |
-| **OpenAI** | GPT-4o, GPT-4 | General purpose |
-| **DeepSeek** | DeepSeek Chat/Coder | Budget-friendly |
-| **Ollama** | Qwen, CodeLlama, etc. | Free, local, private |
+Run `argus init` to configure interactively.
 
----
+### Supported Providers
 
-## API Reference
+| Provider | Cost | Best For |
+|----------|------|----------|
+| **Ollama** | Free (local) | Development, privacy-sensitive code |
+| **ZAI (GLM-4.7)** | ~$3/month | Cost-effective cloud option |
+| **DeepSeek** | Low | Good balance of cost/quality |
+| **Anthropic** | Medium | High-quality analysis |
+| **OpenAI** | Medium | High-quality analysis |
 
-### Worker Service (Port 37778)
+**Note**: The AI provider is only used for `analyze_codebase`. All other tools are completely free.
 
-**Health Check**
-```bash
-curl http://localhost:37778/health
-```
-Response:
-```json
-{
-  "status": "ok",
-  "version": "2.0.0",
-  "cached": 2,
-  "watching": 1
-}
-```
+---
 
-**Load Snapshot**
-```bash
-curl -X POST http://localhost:37778/snapshot/load \
-  -H "Content-Type: application/json" \
-  -d '{"path": "/path/to/.argus/snapshot.txt"}'
-```
+## CLI Reference
 
-**Search**
-```bash
-curl -X POST http://localhost:37778/search \
-  -H "Content-Type: application/json" \
-  -d '{
-    "path": "/path/to/.argus/snapshot.txt",
-    "pattern": "handleAuth",
-    "options": {"caseInsensitive": true, "maxResults": 50}
-  }'
-```
-
-**Get Context**
-```bash
-curl -X POST http://localhost:37778/context \
-  -H "Content-Type: application/json" \
-  -d '{
-    "path": "/path/to/.argus/snapshot.txt",
-    "file": "src/auth.ts",
-    "line": 42,
-    "before": 10,
-    "after": 20
-  }'
-```
+| Command | Description |
+|---------|-------------|
+| `argus init` | Interactive setup wizard |
+| `argus snapshot <path>` | Create a codebase snapshot |
+| `argus analyze <snapshot> <query>` | AI-powered analysis |
+| `argus search <snapshot> <pattern>` | Regex search (no AI) |
+| `argus ui` | Open the web visualizer |
+| `argus mcp install` | Add to Claude Code |
+| `argus status <path>` | Check snapshot freshness |
+| `argus config` | View/edit configuration |
 
 ---
 
-## How It Works
+## FAQ
 
-Argus uses a **Recursive Language Model (RLM)** approach:
+**Q: Do I need to pay for an API?**
 
-1. **Snapshot Creation** - Your codebase is compiled into an optimized text snapshot with structural metadata
-2. **Query Analysis** - The LLM receives your question and the Nucleus DSL reference
-3. **Iterative Exploration** - The LLM generates symbolic commands (grep, filter, map, etc.)
-4. **Command Execution** - Commands run against the full snapshot in a sandbox
-5. **Reasoning Loop** - Results feed back to the LLM for further analysis
-6. **Final Answer** - Once sufficient information is gathered, a comprehensive answer is provided
+No. Use Ollama for free local AI, or just use the zero-cost tools (search, find_importers, etc.) which don't use AI at all.
 
-This allows analysis of codebases **far exceeding** typical context limits (2M+ characters) while using minimal tokens per query.
+**Q: How big can my codebase be?**
 
-### Nucleus DSL Reference
+Argus has been tested on codebases with 100k+ lines. The snapshot file grows linearly (~3KB per 100 lines of code), and search remains fast because it's just regex.
 
-Argus uses the [Nucleus DSL](https://github.com/michaelwhitford/nucleus) for document operations:
+**Q: Should I commit the snapshot to git?**
 
-```lisp
-; Search
-(grep "pattern")                    ; Find matching lines
-(grep "error" "i")                  ; Case-insensitive search
+Optional. Snapshots are deterministic, so you can regenerate them. But committing `.argus/snapshot.txt` means new contributors have it immediately.
 
-; Transform
-(map RESULTS (lambda (x) ...))      ; Transform results
-(filter RESULTS (lambda (x) ...))   ; Filter results
-(sort RESULTS key)                  ; Sort results
+**Q: How do I exclude files?**
 
-; Aggregate
-(count RESULTS)                     ; Count items
-(sum RESULTS)                       ; Sum numeric values
-(first RESULTS)                     ; Get first item
-(take RESULTS n)                    ; Get first n items
-
-; Extract
-(match str "pattern" group)         ; Regex extraction
-(split str delimiter)               ; Split string
-
-; Final Answer
-<<<FINAL>>>your answer here<<<END>>>
+```bash
+argus snapshot . --exclude "node_modules,dist,*.test.ts"
 ```
 
----
-
-## Requirements
-
-- Node.js 18+
-- npm or pnpm
-- API key for your chosen provider (or Ollama for free local usage)
-
----
-
-## FAQ & Documentation
-
-- **[FAQ](./docs/FAQ.md)** - Common questions about costs, workflow, and troubleshooting
-- **[CLAUDE.md Integration](./docs/CLAUDE_MD_INTEGRATION.md)** - How to add Argus to your project's CLAUDE.md
-
-### Quick Answers
+Or add patterns to your config file.
 
-**"Do I need to pay for another API?"**
-No! Use Ollama (free, local) or zero-cost tools like `search_codebase` (no AI at all).
+**Q: Why not just use grep?**
 
-**"I'm starting a fresh project - how does Argus help?"**
-Argus works from Day 0. Snapshot your PRD/TDD, then refresh as you build. See [FAQ](./docs/FAQ.md#im-starting-a-brand-new-project---theres-nothing-to-scan-yet).
-
-**"How do I keep the snapshot updated?"**
-Run `argus status .` to check freshness, then `argus snapshot .` to refresh. See [FAQ](./docs/FAQ.md#how-do-i-keep-the-snapshot-up-to-date).
-
-**"What's the difference between basic and enhanced snapshots?"**
-Enhanced snapshots (default) include import graphs and export indexes that enable zero-cost dependency queries. Basic snapshots are faster to create but only support regex search.
+You can! `search_codebase` is essentially grep. But Argus also gives you:
+- Pre-computed import graph for dependency queries
+- Export index for "where is X defined?"
+- AI analysis when you need deeper understanding
 
 ---
 
 ## Acknowledgments
 
-Argus builds upon and extends the innovative work of [Matryoshka RLM](https://github.com/yogthos/Matryoshka) by [Dmitri Sotnikov (yogthos)](https://github.com/yogthos).
+Argus combines ideas from several projects:
 
-The Matryoshka project introduced the brilliant concept of **Recursive Language Models (RLM)** - using an LLM to generate symbolic commands (via the [Nucleus DSL](https://github.com/michaelwhitford/nucleus)) that are executed against documents, enabling analysis of files far exceeding context window limits. This approach achieves **93% token savings** compared to traditional methods.
+### Matryoshka RLM
 
-**What Argus adds:**
+The core analysis engine is based on [Matryoshka](https://github.com/yogthos/Matryoshka) by [Dmitri Sotnikov (yogthos)](https://github.com/yogthos). It introduced **Recursive Language Models (RLM)** - using an LLM to generate symbolic commands via the [Nucleus DSL](https://github.com/michaelwhitford/nucleus) that navigate large documents. This achieves 93% token savings compared to naive approaches.
 
-| Matryoshka | Argus |
-|------------|-------|
-| Single file analysis | Full codebase analysis |
-| CLI-only | CLI + MCP Server + Web UI |
-| Ollama/DeepSeek providers | Multi-provider (ZAI, Anthropic, OpenAI, Ollama, DeepSeek) |
-| Manual configuration | Interactive setup wizard |
-| Document-focused | Code-aware with structural metadata |
-| No caching | Worker service with LRU cache |
-
-We encourage you to explore the original [Matryoshka](https://github.com/yogthos/Matryoshka) project and the [RLM research paper](https://arxiv.org/abs/2512.24601) that inspired this approach.
+The approach is described in the [RLM research paper](https://arxiv.org/abs/2512.24601).
 
 ### claude-mem
 
-The Argus 2.0 upgrade was heavily inspired by [claude-mem](https://github.com/thedotmack/claude-mem) by [thedotmack](https://github.com/thedotmack). The progressive disclosure architecture, self-documenting MCP tools, and Claude Code integration patterns were adapted from claude-mem's innovative approach to context management.
+The progressive disclosure architecture and Claude Code integration patterns are inspired by [claude-mem](https://github.com/thedotmack/claude-mem) by [thedotmack](https://github.com/thedotmack). The idea of "search first, fetch details only when needed" dramatically reduces token usage.
 
 ---
 
-## License
-
-MIT License - See [LICENSE](./LICENSE)
+## Related Projects
 
-## Contributing
+- [Matryoshka RLM](https://github.com/yogthos/Matryoshka) - The original RLM implementation
+- [Nucleus DSL](https://github.com/michaelwhitford/nucleus) - Symbolic language for document operations
+- [claude-mem](https://github.com/thedotmack/claude-mem) - Progressive disclosure patterns for Claude Code
+- [RLM Paper](https://arxiv.org/abs/2512.24601) - Academic research on Recursive Language Models
 
-Contributions are welcome! Please read our contributing guidelines before submitting PRs.
+---
 
-## Related Projects
+## License
 
-- [Matryoshka RLM](https://github.com/yogthos/Matryoshka) - The original RLM implementation that inspired Argus
-- [claude-mem](https://github.com/thedotmack/claude-mem) - Progressive disclosure and MCP patterns that inspired Argus 2.0
-- [Nucleus DSL](https://github.com/michaelwhitford/nucleus) - The symbolic language used for document operations
-- [RLM Paper](https://arxiv.org/abs/2512.24601) - Academic research on Recursive Language Models
+MIT

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sashabogi/argus-mcp",
-  "version": "2.0.1",
+  "version": "2.0.3",
   "description": "Codebase Intelligence Beyond Context Limits - AI-powered analysis for entire projects",
   "main": "dist/index.js",
   "module": "dist/index.mjs",