sari 0.3.13

Sari - Local Search MCP Server

Sari (사리)

🇰🇷 한국어 가이드 (Korean Guide)

Sari is a high-performance Local Code Search Agent implementing the Model Context Protocol (MCP). It empowers AI assistants (like Claude, Cursor, Codex) to efficiently navigate, understand, and search large codebases without sending code to external servers.

Key Features:

• ⚡ Fast Indexing: SQLite FTS5 + AST-based symbol extraction.
• 🔍 Smart Search: Hybrid ranking (Keyword + Symbol structure).
• 🧠 Code Intelligence: Call graphs, snippets management, and domain context archiving.
• 🔒 Local & Secure: All data remains on your machine. No external API dependency.

---

---

🚀 Installation & Setup

Choose the method that best fits your workflow. Sari is extremely lightweight (< 5MB) by default.

Method 1: Automatic Script (Recommended)

This script handles everything, including binary detection and interactive feature selection. It will automatically use uv if available for 10x faster installation.

🍎 macOS / Linux

curl -fsSL https://raw.githubusercontent.com/BaeCheolHan/sari/main/install.py | python3 - -y --update

🪟 Windows (PowerShell)

irm https://raw.githubusercontent.com/BaeCheolHan/sari/main/install.py | python - -y --update

---

Method 2: Modern CLI Setup (via uv)

For power users who want a clean, isolated installation with automatic PATH management.

# Recommended: Install as a global tool
uv tool install sari

# Install with all high-precision features (CJK + Tree-sitter)
uv tool install "sari[full]"

# Or run instantly without installation
uv x sari status

---

Method 3: Legacy Installation (via pip)

Standard installation for environments without uv.

# Core only
pip install sari

# Core + CJK + Tree-sitter
pip install "sari[full]"

---

Method 4: HTTP Mode (Most Stable)

If you experience Connection closed errors with stdio, use HTTP mode. This separates the server process from the CLI noise.

1. Start Sari HTTP Server with Environment Variables:

# Set workspace and log level inline
SARI_WORKSPACE_ROOT=/absolute/path/to/project \
SARI_LOG_LEVEL=INFO \
sari --transport http --http-api-port 47777 --http-daemon

2. Client Configuration:

Gemini CLI (~/.gemini/settings.json)

{
  "mcpServers": {
    "sari": {
      "url": "http://127.0.0.1:47777/mcp"
    }
  }
}

Codex CLI (.codex/config.toml)

[mcp_servers.sari]
url = "http://127.0.0.1:47777/mcp"
enabled = true

---

🏎️ Optional Features (Selectable Extras)

Sari allows you to choose between low footprint and high precision.

Extra	Feature	Approx. Size	Installation
Core	Standard Regex parsing, FTS5 Search	< 5MB	pip install sari
[cjk]	Accurate KR/JP/CN Tokenization	+50MB	pip install "sari[cjk]"
[treesitter]	High-precision AST Symbol extraction	+10MB~	pip install "sari[treesitter]"
[full]	All of the above + Tantivy Engine	+100MB+	pip install "sari[full]"

Verification

After installation, verify your active features:

sari doctor
# If 'sari' command is not found, use:
# python3 -m sari doctor

---

🔌 Client Configuration

To use Sari with your AI assistant, add it to your MCP configuration file.

1. Gemini CLI / Codex CLI

File: ~/.gemini/settings.json (or .codex/config.toml)

{
  "mcpServers": {
    "sari": {
      "command": "/Users/YOUR_USERNAME/.local/bin/sari",
      "args": ["--transport", "stdio", "--format", "pack"],
      "env": {
        "SARI_WORKSPACE_ROOT": "/absolute/path/to/your/project"
      }
    }
  }
}

Note: Use the absolute path to the sari executable (usually in ~/.local/bin/sari) to ensure it runs even if not in your system PATH.

2. Claude Desktop & Cursor

{
  "mcpServers": {
    "sari": {
      "command": "/Users/YOUR_USERNAME/.local/bin/sari",
      "args": ["--transport", "stdio", "--format", "pack"],
      "env": {
        "SARI_WORKSPACE_ROOT": "/absolute/path/to/your/project",
        "SARI_RESPONSE_COMPACT": "1"
      }
    }
  }
}

3. Claude Code (CLI)

Run the following command:

claude mcp add sari -- uv tool run sari --transport stdio --format pack

---

⚙️ Configuration Reference

Variables are categorized into Installation-time and Runtime settings.

How to set environment variables

• MCP Client: Add to the env block of your MCP server configuration.
• CLI: Prefix the command, e.g., SARI_ENGINE_MODE=sqlite sari status.
• Permanent: Export in your shell profile (e.g., ~/.zshrc).

"env": {
  "SARI_WORKSPACE_ROOT": "/path/to/project",
  "SARI_LOG_LEVEL": "ERROR",
  "SARI_ENGINE_TOKENIZER": "cjk"
}

A. Installation & Bootstrapping

Settings affecting the installation scripts (install.py, bootstrap.sh).

Variable	Description	Default
XDG_DATA_HOME	Custom data directory for installation. Sari installs to $XDG_DATA_HOME/sari.	~/.local/share
SARI_SKIP_INSTALL	Set 1 to skip automatic pip install/upgrade on startup when using the bootstrap script. Useful for development or offline usage.	0
SARI_NO_INTERACTIVE	Set 1 to disable interactive prompts during installation (assumes 'yes').	0

B. System & Runtime

Settings controlling the MCP server loop and behaviors. Add these to your env config.

1. Core & System

Essential settings for basic operation. (SARI_ prefix is also supported for backward compatibility).

Variable	Description	Default
SARI_WORKSPACE_ROOT	(Required) Absolute path to the project root. Auto-detected if omitted.	Auto-detect
SARI_ROOTS_JSON	JSON array of strings for multiple workspace roots. e.g., ["/path/a", "/path/b"]	-
SARI_DB_PATH	Custom path for the SQLite database file.	~/.local/share/sari/index.db
SARI_CONFIG	Path to a specific config file to load.	~/.config/sari/config.json
SARI_DATA_DIR	Override global data directory for DB, engine, and caches.	~/.local/share/sari
SARI_RESPONSE_COMPACT	Minify JSON responses (pack format) to save LLM tokens. Set 0 for pretty-print debugging.	1 (Enabled)
SARI_FORMAT	Output format for CLI tools. pack (text-based) or json.	pack

2. Search Engine

Settings to tune search quality and backend behavior.

Variable	Description	Default
SARI_ENGINE_MODE	Search backend. embedded uses Tantivy (faster, smart ranking), sqlite uses FTS5 (slower, fallback).	embedded
SARI_ENGINE_TOKENIZER	Tokenizer strategy. auto (detects), cjk (optimized for KR/CN/JP), latin (standard).	auto
SARI_ENGINE_AUTO_INSTALL	Automatically install engine binaries (Tantivy) if missing.	1 (Enabled)
SARI_ENGINE_SUGGEST_FILES	File count threshold to suggest upgrading to Tantivy engine in status checks.	10000
SARI_LINDERA_DICT_PATH	Path to custom Lindera dictionary for CJK tokenization (Advanced).	-
SARI_ENGINE_MEM_MB	Total embedded engine memory budget (MB).	512
SARI_ENGINE_INDEX_MEM_MB	Embedded engine indexing memory budget (MB).	256
SARI_ENGINE_THREADS	Embedded engine thread count.	2
SARI_ENGINE_MAX_DOC_BYTES	Max document bytes to index in engine.	4194304
SARI_ENGINE_PREVIEW_BYTES	Preview bytes per document.	8192

Config file equivalents (config.json):

{
  "engine_mode": "embedded",
  "engine_auto_install": true
}

SARI_ENGINE_MODE and SARI_ENGINE_AUTO_INSTALL override these values at runtime.

3. Indexing & Performance

Fine-tune resource usage and concurrency.

Variable	Description	Default
SARI_COALESCE_SHARDS	Number of lock shards for indexing concurrency. Increase for massive repos with frequent changes.	16
SARI_PARSE_TIMEOUT_SECONDS	Timeout per file parsing in seconds. Set 0 to disable timeout. Prevents parser hangs.	0
SARI_PARSE_TIMEOUT_WORKERS	Worker threads for parsing with timeout.	2
SARI_MAX_PARSE_BYTES	Max file size to attempt parsing (bytes). Larger files are skipped or sampled.	16MB
SARI_MAX_AST_BYTES	Max file size to attempt AST extraction (bytes).	8MB
SARI_GIT_CHECKOUT_DEBOUNCE	Seconds to wait after git checkout before starting bulk indexing.	3.0
SARI_FOLLOW_SYMLINKS	Follow symbolic links during file scanning. Caution: May cause infinite loops if circular links exist.	0 (Disabled)
SARI_MAX_DEPTH	Maximum directory depth to scan. Prevents infinite loops.	30
SARI_READ_MAX_BYTES	Max bytes returned by read_file tool. Prevents context overflow.	1MB
SARI_INDEX_MEM_MB	Overall indexing memory budget (MB).	512
SARI_INDEX_WORKERS	Override index worker count.	2
SARI_AST_CACHE_ENTRIES	LRU cache size for Tree-sitter ASTs.	128

4. Network & Security

Connectivity settings for the daemon.

Variable	Description	Default
SARI_DAEMON_HOST	Host address for the background daemon.	127.0.0.1
SARI_DAEMON_PORT	TCP port for the daemon.	47779
SARI_HTTP_API_PORT	Port for the HTTP API server (optional).	47777
SARI_ALLOW_NON_LOOPBACK	Allow connections from non-localhost IPs. Security Risk: Only enable in trusted networks.	0 (Disabled)

5. Advanced / Debug

Developer options for debugging and plugin extension.

Variable	Description	Default
SARI_LOG_LEVEL	Logging verbosity (DEBUG, INFO, WARNING, ERROR).	INFO
SARI_DRYRUN_LINT	Enable syntax checking (linting) in dry-run-diff.	0 (Disabled)
SARI_PERSIST_ROOTS	Set 1 to persist detected roots to config.json.	0 (Disabled)
SARI_CALLGRAPH_PLUGIN	Python module path for custom static analysis plugin.	-
SARI_DLQ_POLL_SECONDS	Interval to retry failed indexing tasks (Dead Letter Queue).	60

---

🛠️ Usage (MCP Tools)

Once connected, your AI assistant can use these tools:

Core Tools

• search: Search for code or documentation using keywords or regex.
• read_file: Read file content (optimized for large files).
• list_files: List files in the repository.
• search_symbols: Find classes, functions, or methods by name.
• read_symbol: Read only the definition of a specific symbol (saves context).

Intelligence Tools

• call_graph: Analyze function call relationships (upstream/downstream).
• save_snippet / get_snippet: Save and retrieve important code blocks with tags.
• archive_context / get_context: Store domain knowledge and design decisions.
• grep_and_read: Search and read top N files in one go (Composite tool).

---

🩺 Troubleshooting

Check Status

You can check the daemon status and indexing progress:

sari status

sari status will automatically use the actual HTTP port recorded in .codex/tools/sari/data/server.json (workspace-local). The daemon port is discovered via the global registry at ~/.local/share/sari/server.json, so clients can reconnect without manual port changes.

If Daemon Port Is Busy

If you see a message like "Daemon already running" but things still don't work, another process may be using the default port.

# Try a different daemon port:
SARI_DAEMON_PORT=47790 sari daemon start -d

Run Daemon + HTTP Together

sari status talks to the HTTP server, so you should run the daemon and HTTP together. The daemon auto-starts HTTP for the current workspace.

# Start both (daemon will auto-start HTTP):
sari daemon start -d

# If you need a custom workspace:
SARI_WORKSPACE_ROOT=/path/to/workspace sari daemon start -d

Zero-Downtime Upgrade (Port Split)

You can run a new daemon+HTTP on different ports, switch clients, then stop the old one.

# Start new instance on alternate ports:
sari daemon start -d --daemon-port 47790 --http-port 47778

# Check new instance:
sari status --daemon-port 47790 --http-port 47778

Run Doctor

Diagnose issues with your environment or installation:

sari doctor --auto-fix

Update

Update Sari using the installer script:

curl -fsSL https://raw.githubusercontent.com/BaeCheolHan/sari/main/install.py | python3 - --update -y

After updating, restart the daemon to load the new version:

pkill -f "sari.mcp.daemon"
sari daemon start -d

The bootstrap script now starts a new daemon on a free port automatically to allow zero-downtime updates.

Storage Maintenance

To prevent unlimited growth of auxiliary data (snippets, error logs, etc.), Sari implements TTL (Time-To-Live) policies. Existing data is automatically cleaned up based on TTL settings, or you can manually trigger it.

Manual Prune:

# Prune all tables using default/configured TTL
sari prune

# Prune specific table with custom days
sari prune --table failed_tasks --days 3

TTL Configuration (Environment Variables):

• SARI_STORAGE_TTL_DAYS_SNIPPETS (Default: 30)
• SARI_STORAGE_TTL_DAYS_FAILED_TASKS (Default: 7)
• SARI_STORAGE_TTL_DAYS_CONTEXTS (Default: 30)

Uninstall

To remove Sari, indexed data, and default configs:

# macOS/Linux
curl -fsSL https://raw.githubusercontent.com/BaeCheolHan/sari/main/install.py | python3 - --uninstall

# Windows
irm https://raw.githubusercontent.com/BaeCheolHan/sari/main/install.py | python - --uninstall

To also remove workspace-local caches (if used), pass the workspace root:

curl -fsSL https://raw.githubusercontent.com/BaeCheolHan/sari/main/install.py | python3 - --uninstall --workspace-root /path/to/project

The uninstall command also scans your home directory for .codex/tools/sari caches and removes them (best effort).

If you set SARI_CONFIG or SARI_CONFIG to a custom path and want that file removed too, pass:

curl -fsSL https://raw.githubusercontent.com/BaeCheolHan/sari/main/install.py | python3 - --uninstall --force-config

---

📜 License

Apache License 2.0