# CodeMemoryPrime (cmp)

> CodeMemoryPrime (`cmp`) is a fast, local Rust Model Context Protocol (MCP) codebase intelligence server that cuts AI API token costs by 90%+ and prevents context loss in agentic coding workflows.

## Key Facts & Overview
- **Name**: CodeMemoryPrime (CLI binary: `cmp`)
- **Developer**: Rickie Blevins
- **Primary Function**: Local vector RAG and persistent memory for AI coding assistants (Claude Desktop, Cursor IDE, Windsurf, Google Antigravity).
- **Token Efficiency**: Replaces 100,000+ token full-file context dumps with RRF hybrid search (Cosine Vector + FTS5 keyword indexing), retrieving only relevant functions and parent AST struct headers.
- **Performance**: Sub-0.2 second re-indexing via SHA-256 sub-chunk caching for untouched files.
- **Privacy & Execution**: 100% local execution using Ollama (`qwen2.5-coder:7b`, `nomic-embed-text`) or LM Studio endpoints. Zero data telemetry.
- **Git Safety Checkpoints**: Bookmarks repository state before AI refactoring operations (`create_checkpoint`), enabling instant rollback if prompts stray off course.

## Supported MCP Clients
- Claude Desktop (`claude_desktop_config.json`)
- Cursor IDE (`~/.cursor/mcp.json`)
- Codeium Windsurf (`~/.codeium/windsurf/mcp_config.json`)
- Google Antigravity AGY (`~/.gemini/antigravity/mcp_config.json`)

## Licensing Terms
- **BSL 1.1 (Business Source License)**: Free forever for personal use, open-source maintainers, students, hobbyists, and small teams (<= 3 developers AND < $100,000 annual revenue).
- **Commercial Enterprise License**: $15/seat/month for organizations with > 3 developers or >= $100,000 revenue. Offline Ed25519 seat keys.
- **Commercial Inquiries**: license@rickieblevins.com
- **Repository**: https://github.com/kichikubeihei/CodeMemoryPrime
- **Documentation**: https://codememoryprime.com/

## 35 MCP Tool Suite Reference
CodeMemoryPrime exposes 35 Model Context Protocol tools over stdio JSON-RPC:

### 1. Search & Indexing
- `index_workspace`: Recursively indexes codebases into local SQLite vector memory using SHA-256 sub-chunk caching.
- `search_codebase`: Performs RRF hybrid search attaching parent struct headers and imports to code snippets to cut LLM token costs.
- `get_dependencies`: Retrieves module dependency graphs, import statements, and caller/callee relationships.

### 2. Persistent Memory
- `save_interaction`: Records architectural decisions, preferences, and guidelines into persistent SQLite memory logs.
- `search_memories`: Searches consolidated architectural facts and interaction journals using vector + FTS search.
- `consolidate_memories`: Runs background consolidation over raw logs, distilling them into structured permanent facts.

### 3. File Management
- `read_file`: Reads text files with line range slice notation (start_line, end_line).
- `write_file`: Writes or overwrites text files with automatic parent directory creation.
- `patch_file`: Applies targeted search-and-replace snippet edits to files.
- `list_files`: Lists directory contents recursively or flat with file metadata.

### 4. Git & Checkpoints
- `run_command`: Executes shell commands within a safe allowlist (cargo, npm, git, cmp, etc.).
- `git`: Unified Git CLI wrapper supporting status, diff, log, branch, commit, and checkout.
- `create_checkpoint`: Bookmarks Git state before major edits so you can roll back instantly if things go wrong.
- `restore_checkpoint`: Restores a previously saved Git safety checkpoint.
- `list_checkpoints`: Lists all saved Git safety checkpoints for a project.

### 5. Refactoring & Diagnostics
- `explain_code`: Generates multi-level code explanations (high_level, line_by_line, architecture).
- `refactor_code`: Applies targeted code transformations (clean, extract_function, add_types, convert_async).
- `review_code`: Multi-perspective code review covering readability, performance bottlenecks, and security.
- `check_security`: Audits code snippets for OWASP top vulnerabilities, memory safety flaws, and secret leaks.
- `optimize_code`: Analyzes code for memory efficiency, execution speed, and algorithmic complexity.
- `generate_tests`: Generates unit test suites using cargo test, pytest, or jest.
- `diagnose_compiler_error`: Analyzes compiler or build error outputs and proposes direct fixes.

### 6. Docs & Framework RAG
- `index_framework_specifications`: Indexes external framework docs or Markdown specs into vector memory.
- `search_framework_specifications`: Performs hybrid search across indexed framework specs and API docs.
- `generate_documentation`: Generates docstrings and inline technical documentation for code symbols.
- `get_documentation`: Retrieves extracted symbol documentation comments and AST docstrings.

### 7. Modularization & Analytics
- `modularize_code`: Proposes microservice or plugin refactoring boundaries for large monolithic components.
- `extract_plugin`: Extracts standalone plugin definitions and export interfaces from code.
- `publish_plugin`: Publishes plugin definitions into the local catalog FTS database.
- `recommend_plugins`: Recommends local catalog plugins based on project requirements.
- `log_token_usage`: Logs token usage and savings vs full context window dumps for LLM queries.
- `get_token_analytics`: Retrieves total token reduction metrics and efficiency percentages for a project.

### 8. Health & System
- `project_health`: Reports SQLite DB size, Ollama/OpenAI connection status, auto-detected models, and license status.
- `summarize_project`: Synthesizes detailed architectural briefs and structural summaries for indexed projects.
- `configure_settings`: Configures LLM provider, base URL, generation/embedding models, or API keys.