Snipara MCP

Model Context Protocol (MCP) server for Snipara - Context optimization and Agent infrastructure for LLMs.

Overview

Snipara MCP is a client package that connects AI assistants to the Snipara API, enabling:

• 🔍 90% Context Reduction - Query 500K tokens, get 5K relevant results
• 🌐 Multi-Project Search - Search across all your repos in one query
• 🧠 AI Memory - Persistent semantic memory across sessions
• 👥 Team Standards - Auto-inject coding standards into every query
• 🤖 Multi-Agent Swarms - Coordinate multiple AI agents with shared state

How It Works

┌─────────────────────────────────────────────────────────────────┐
│  Your AI Assistant (Claude, GPT, Cursor, etc.)                 │
│  ├── Asks: "How does authentication work?"                     │
│  └── Uses: rlm_context_query("authentication")                 │
└────────────────────────────┬────────────────────────────────────┘
                             │
                   ┌─────────▼─────────┐
                   │  Snipara MCP      │  (This package)
                   │  Client Library   │
                   └─────────┬─────────┘
                             │ HTTPS
                             ▼
                   ┌─────────────────────┐
                   │  Snipara API        │
                   │  api.snipara.com    │
                   │  - Indexed docs     │
                   │  - Semantic search  │
                   │  - Agent memory     │
                   └─────────┬───────────┘
                             │
                   ┌─────────▼─────────┐
                   │  Returns: 3K      │
                   │  relevant tokens  │
                   │  instead of 50K   │
                   └───────────────────┘

Key Insight: This is the client package - it connects to Snipara's API to fetch optimized context. The server-side implementation is private.

What's In This Repository

This is the open-source MCP client package published to PyPI as snipara-mcp.

snipara-mcp/
├── src/snipara_mcp/
│   ├── __init__.py         # Package entry point
│   ├── server.py           # MCP stdio transport (2,050 lines)
│   ├── rlm_tools.py        # 39 tool implementations (2,117 lines)
│   └── auth.py             # OAuth device flow (402 lines)
├── skill.md                # ClawdHub marketplace documentation
├── pyproject.toml          # PyPI package configuration
└── README.md               # This file

What's NOT in this repo (private):

• FastAPI backend server
• Database models and search algorithms
• Rate limiting and billing logic
• Internal API endpoints

Quick Start

Installation

Option 1: uvx (No Install Required)

uvx snipara-mcp

Option 2: pip

pip install snipara-mcp

Option 3: With RLM Runtime

pip install snipara-mcp[rlm]

Configuration

Get your API key from snipara.com/dashboard

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "snipara": {
      "command": "uvx",
      "args": ["snipara-mcp"],
      "env": {
        "SNIPARA_API_KEY": "rlm_your_key",
        "SNIPARA_PROJECT_ID": "your_project_id"
      }
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "snipara": {
      "command": "uvx",
      "args": ["snipara-mcp"],
      "env": {
        "SNIPARA_API_KEY": "rlm_your_key",
        "SNIPARA_PROJECT_ID": "your_project_id"
      }
    }
  }
}

Claude Code

claude mcp add snipara -- uvx snipara-mcp
export SNIPARA_API_KEY="rlm_your_key"
export SNIPARA_PROJECT_ID="your_project_id"

Windsurf

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

{
  "mcpServers": {
    "snipara": {
      "command": "uvx",
      "args": ["snipara-mcp"],
      "env": {
        "SNIPARA_API_KEY": "rlm_your_key",
        "SNIPARA_PROJECT_ID": "your_project_id"
      }
    }
  }
}

Your First Query

Once configured, ask your AI:

"Use snipara to find how authentication works"

Behind the scenes:

rlm_context_query("authentication")
→ 2 seconds later
→ Returns top 3 relevant docs (3K tokens instead of 50K)

Features

🎯 Context Optimization (All Plans)

Tools: rlm_context_query, rlm_ask, rlm_search

Query your documentation with semantic search and get only relevant results within your token budget.

{
  "query": "authentication implementation",
  "max_tokens": 6000,
  "search_mode": "hybrid"
}

🌐 Multi-Project Search (Team+)

Tool: rlm_multi_project_query

Search across all your team's repositories in one query.

{
  "query": "How do we handle webhooks?",
  "project_ids": [],
  "max_tokens": 8000
}

🧠 AI Memory (Agents Plans)

Tools: rlm_remember, rlm_recall

Persistent semantic memory across sessions with confidence decay.

// Store once
{
  "content": "User prefers TypeScript strict mode",
  "type": "preference",
  "scope": "project"
}
 
// Recall later
{
  "query": "What are my coding preferences?",
  "limit": 5
}

👥 Team Standards (Team+)

Tool: rlm_shared_context

Auto-inject team coding standards into every query.

{
  "categories": ["MANDATORY", "BEST_PRACTICES"],
  "max_tokens": 4000
}

🤖 Multi-Agent Swarms (Enterprise)

Tools: rlm_swarm_create, rlm_claim, rlm_task_create, etc.

Coordinate multiple AI agents with:

• Shared state management
• Resource claims (prevent conflicts)
• Distributed task queue
• Event broadcasting

// Create swarm
{ "name": "code-review-swarm", "max_agents": 5 }
 
// Claim file access
{ "swarm_id": "...", "resource_type": "file", "resource_id": "src/auth.ts" }
 
// Create task
{ "swarm_id": "...", "title": "Review auth module", "priority": 90 }

Complete Tool Reference

Primary Tools

• rlm_context_query - Semantic search with token budget
• rlm_ask - Quick keyword search
• rlm_search - Regex pattern search

Advanced Tools

• rlm_multi_query - Parallel queries
• rlm_decompose - Break down complex questions
• rlm_plan - Generate execution plan
• rlm_inject - Set session context

Team Tools

• rlm_multi_project_query - Cross-repo search
• rlm_shared_context - Team standards
• rlm_list_templates - Prompt templates
• rlm_upload_shared_document - Upload team docs

Memory Tools (Agents Plan)

• rlm_remember - Store memory
• rlm_recall - Semantic recall
• rlm_memories - List all memories
• rlm_forget - Delete memories

Swarm Tools (Enterprise)

• rlm_swarm_create / rlm_swarm_join
• rlm_claim / rlm_release
• rlm_state_get / rlm_state_set
• rlm_broadcast
• rlm_task_create / rlm_task_claim / rlm_task_complete

Document Management

• rlm_upload_document - Upload single doc
• rlm_sync_documents - Bulk sync
• rlm_store_summary - Store LLM-generated summary
• rlm_stats - Documentation statistics

Total: 39 tools

Full documentation: docs.snipara.com

Pricing

Context Plans (Documentation Search)

Agents Plans (Memory & Swarms)

Note: Context and Agents are separate subscriptions.

Development

Prerequisites

• Python 3.10+
• pip or uv

Local Setup

# Clone the repo
git clone https://github.com/alopez3006/snipara-mcp.git
cd snipara-mcp
 
# Install dependencies
pip install -e .
 
# Set environment variables
export SNIPARA_API_KEY="rlm_your_key"
export SNIPARA_PROJECT_ID="your_project_id"
 
# Test the package
python -m snipara_mcp

Project Structure

src/snipara_mcp/
├── __init__.py         # Package entry point, CLI main()
├── server.py           # MCP stdio transport implementation
│                       # - Handles MCP protocol (JSON-RPC)
│                       # - Tool registration and dispatch
│                       # - Prompt auto-injection
│
├── rlm_tools.py        # 39 tool implementations
│                       # - Context optimization tools
│                       # - Agent memory tools
│                       # - Multi-agent swarm tools
│                       # - Document management
│
└── auth.py             # OAuth device flow authentication
                        # - Login/logout/status CLI commands
                        # - Token storage and refresh

Testing

# Test basic connection
uvx snipara-mcp
 
# Test with specific tool
echo '{"query": "test"}' | uvx snipara-mcp
 
# Test OAuth flow
snipara-mcp-login
snipara-mcp-status

Publishing to PyPI

# Bump version in pyproject.toml
# version = "2.2.1"
 
# Build distribution
python -m build
 
# Upload to PyPI
python -m twine upload dist/*

Architecture

MCP Protocol

This package implements the Model Context Protocol stdio transport:

AI Client → stdin → snipara-mcp → HTTPS → api.snipara.com
AI Client ← stdout ← snipara-mcp ← JSON ← api.snipara.com

Authentication

Supports two authentication methods:

1. API Key (Recommended)

   export SNIPARA_API_KEY="rlm_..."

2. OAuth Device Flow

   snipara-mcp-login
   # Follow browser prompt to authorize

Tokens are stored in ~/.snipara/credentials.json

Rate Limiting

Rate limits are enforced server-side based on your plan:

• FREE: 100 queries/month
• PRO: 5,000 queries/month
• TEAM: 20,000 queries/month
• ENTERPRISE: Unlimited

Upgrading

When a new version is released:

1. Clear uvx cache

# macOS/Linux
rm -rf ~/.cache/uv/tools/snipara-mcp
rm -rf ~/Library/Caches/uv/tools/snipara-mcp
 
# Windows
rmdir /s %LOCALAPPDATA%\uv\tools\snipara-mcp

2. Restart your MCP client

MCP tool definitions are loaded at startup. Restart Claude Desktop, Cursor, or your MCP client.

3. Verify version

snipara-mcp-status

Troubleshooting

MCP tools not showing up

1. Restart your MCP client - Tools are cached at startup
2. Clear uvx cache - See Upgrading section
3. Check config syntax - Ensure valid JSON in MCP config

"Invalid API key" error

• Verify API key in dashboard
• Check for whitespace in config
• Try OAuth: snipara-mcp-login

MCP server not connecting

• Check uvx is installed: uvx --version
• Install uv: curl -LsSf https://astral.sh/uv/install.sh | sh
• Check Claude Code output panel for errors

RLM Runtime Integration

Use Snipara tools programmatically in Python:

from rlm import RLM
 
# Snipara tools auto-registered when credentials set
rlm = RLM(
    model="claude-sonnet-4-20250514",
    snipara_api_key="rlm_your_key",
    snipara_project_slug="your-project"
)
 
# LLM can now query your docs during execution
result = rlm.run("Implement auth following our coding standards")

See: rlm-runtime documentation

CI/CD Integration

Auto-sync docs on git push:

curl -X POST "https://api.snipara.com/v1/YOUR_PROJECT/webhook/sync" \
  -H "X-API-Key: YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"documents": [{"path": "docs/api.md", "content": "..."}]}'

See: GitHub Action example (coming soon)

Support

• Website: snipara.com
• Documentation: docs.snipara.com
• Issues: GitHub Issues
• Email: support@snipara.com
• Discord: Join our community

Contributing

We welcome contributions! Please:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

Development Guidelines:

• Follow existing code style
• Add docstrings to public functions
• Update README if adding features
• Test with multiple MCP clients

Version History

License

MIT License - see LICENSE file

Built with ❤️ by the Snipara team

Website • Documentation • PyPI • GitHub