Write effective MCP server instructions for tool compaction support. Use when creating or updating the `instructions` parameter in MaaSProtectedFastMCP server initialization. Helps LLMs understand when to search for tools, workflow patterns, and cross-feature relationships.
Add this skill
npx mdskills install adawalli/mcp-tool-skillComprehensive guide for writing workflow-focused MCP server instructions with clear templates and evidence
AI integration without the complexity
Intelligent AI model search and discovery with zero-install simplicity
Quick Start โข Features โข Documentation โข Contributing
Nexus is a Model Context Protocol (MCP) server that provides AI-powered search functionality through the OpenRouter API. It integrates with MCP-compatible clients including Claude Desktop and Cursor, providing search capabilities via multiple model families including Perplexity Sonar (real-time web search) and Grok 4 (training-data knowledge).
bunx (or npx) with no build requirementssonar - Fast Q&A, real-time web search (30s timeout, standard tier)sonar-pro - Multi-step queries, real-time web search (60s timeout, premium tier)sonar-reasoning-pro - Chain-of-thought reasoning, real-time web search (120s timeout, premium tier)sonar-deep-research - Exhaustive research reports, real-time web search (300s timeout, premium tier)grok-4 - Training-data knowledge, no real-time search (60s timeout, premium tier)Execute the server without local installation:
# Set your OpenRouter API key
export OPENROUTER_API_KEY=your-api-key-here
# Run the server via bunx (recommended)
bunx nexus-mcp
# Or via npx
npx nexus-mcp
The server starts and listens for MCP client connections via STDIO transport.
# Test the CLI help
bunx nexus-mcp --help
# Test the version
bunx nexus-mcp --version
# Run with your API key
OPENROUTER_API_KEY=your-key bunx nexus-mcp
For local development or customization:
git clone https://github.com/adawalli/nexus.git
cd nexus
bun install
bun run build
# Copy the example environment file
cp .env.example .env
# Edit .env and add your actual API key
# OPENROUTER_API_KEY=your-api-key-here
bun run start
Configure MCP clients to execute the server via bunx:
Configuration in ~/.claude/mcp_settings.json:
{
"mcpServers": {
"nexus": {
"command": "bunx",
"args": ["nexus-mcp"],
"env": {
"OPENROUTER_API_KEY": "your-api-key-here"
}
}
}
}
Restart Claude Code after configuration changes.
Add server configuration in Cursor's MCP settings:
nexusbunx["nexus-mcp"]OPENROUTER_API_KEY=your-api-key-hereRestart Cursor after configuration changes.
Standard MCP client connection parameters:
bunx["nexus-mcp"]OPENROUTER_API_KEY=your-api-key-hereIf you don't have Bun installed, use npx in place of bunx in any of the configurations above.
For a local installation (after following the local development setup):
{
"mcpServers": {
"nexus": {
"command": "bun",
"args": ["run", "/path/to/nexus-mcp/dist/cli.js"],
"env": {
"OPENROUTER_API_KEY": "your-api-key-here"
}
}
}
}
Once integrated, you can use the search tool in your MCP client:
Use the search tool to find information about "latest developments in AI"
Search for "climate change solutions" using:
- Model: sonar-pro
- Max tokens: 2000
- Temperature: 0.3
# Fast Q&A with real-time web search (default)
Search for "latest news" with model: sonar
# Deep research with comprehensive analysis
Search for "AI safety research" with model: sonar-deep-research
# Knowledge from training data (no web search)
Search for "explain quantum computing" with model: grok-4
searchThe main search tool that provides AI-powered search capabilities.
Parameters:
query (required): Search query (1-2000 characters)model (optional): Model to use (default: sonar)
sonar - Fast Q&A with real-time web search (30s timeout)sonar-pro - Multi-step queries with real-time web search (60s timeout, premium)sonar-reasoning-pro - Chain-of-thought reasoning with real-time web search (120s timeout, premium)sonar-deep-research - Exhaustive research reports with real-time web search (300s timeout, premium)grok-4 - Training-data knowledge, no real-time search (60s timeout, premium)maxTokens (optional): Maximum response tokens (1-4000, default: 1000)temperature (optional): Response randomness (0-2, default: 0.3)timeout (optional): Override default timeout in milliseconds (5000-600000)Example Response (Perplexity model):
Based on current information, here are the latest developments in AI...
[Detailed AI-generated response with current information]
---
**Search Metadata:**
- Model: perplexity/sonar
- Response time: 1250ms
- Tokens used: 850
- Timeout: 30000ms
- Search type: realtime
- Sources: 5 found
Example Response (Grok 4 model):
Quantum computing is a type of computation that harnesses quantum mechanics...
[Response based on training data knowledge]
---
**Search Metadata:**
- Model: x-ai/grok-4
- Response time: 3500ms
- Tokens used: 650
- Timeout: 60000ms
- Search type: training-data
- Cost tier: premium
OPENROUTER_API_KEY (required): Your OpenRouter API keyNODE_ENV (optional): Environment setting (development, production, test)LOG_LEVEL (optional): Logging level (debug, info, warn, error)The server supports additional configuration through environment variables:
OPENROUTER_TIMEOUT_MS: Request timeout in milliseconds (default: 30000)OPENROUTER_MAX_RETRIES: Maximum retry attempts (default: 3)OPENROUTER_BASE_URL: Custom OpenRouter API base URLThe server provides a configuration status resource at config://status that shows:
"bunx: command not found"
curl -fsSL https://bun.sh/install | bash"npx: command not found"
node --versionnpm install -g npm@latest"Cannot find package 'nexus-mcp'"
Slow startup on first run
"Permission denied" errors with npx
npx --yes nexus-mcp --stdionpm config set user 0 && npm config set unsafe-perm true"Search functionality is not available"
OPENROUTER_API_KEY environment variable is set"Authentication failed: Invalid API key"
"Rate limit exceeded"
Connection timeouts
OPENROUTER_TIMEOUT_MS=60000MCP client can't connect to server
Enable debug logging by:
For local development: Add LOG_LEVEL=debug to your .env file
For MCP clients: Add LOG_LEVEL: "debug" to the env section of your MCP configuration
This will provide detailed information about:
You can test if the server is working by checking the configuration status resource in your MCP client, or by running a simple search query.
For developers working on this server:
# Development with hot reload
bun run dev
# Run tests
bun run test
# Run tests with coverage
bun run test:coverage
# Lint code
bun run lint
# Format code
bun run format
OpenRouter charges for API usage based on token consumption:
| ๐ Guide | ๐ Link | ๐ Description |
|---|---|---|
| Quick Start | Getting Started | Zero-install setup in 30 seconds |
| API Reference | MCP Tools | Complete command reference |
| Configuration | Environment Setup | Advanced configuration options |
| Contributing | Contributing Guide | Join our open source community |
| Troubleshooting | Common Issues | Solutions to common problems |
We welcome contributions from developers of all experience levels!
Contributors are recognized in our:
| ๐ฌ Need Help? | ๐ Resource |
|---|---|
| Quick Questions | GitHub Discussions |
| Bug Reports | GitHub Issues |
| Documentation | OpenRouter Docs โข MCP Specification |
| Feature Requests | Enhancement Proposals |
MIT License - see LICENSE file for details.
Made with โค๏ธ by the open source community
โญ Star us on GitHub โข ๐ฆ View on NPM โข ๐ Read the Docs
Nexus: AI integration without the complexity
Best experience: Claude Code
/plugin marketplace add adawalli/mcp-tool-skillThen /plugin menu โ select skill โ restart. Use /skill-name:init for first-time setup.
Other platforms
Install via CLI
npx mdskills install adawalli/mcp-tool-skillMCP Server Instructions is a free, open-source AI agent skill. Write effective MCP server instructions for tool compaction support. Use when creating or updating the `instructions` parameter in MaaSProtectedFastMCP server initialization. Helps LLMs understand when to search for tools, workflow patterns, and cross-feature relationships.
Install MCP Server Instructions with a single command:
npx mdskills install adawalli/mcp-tool-skillThis downloads the skill files into your project and your AI agent picks them up automatically.
MCP Server Instructions works with Claude Code, Claude Desktop, Cursor, Vscode Copilot, Windsurf, Continue Dev, Codex, Gemini Cli, Amp, Roo Code, Goose, Opencode, Trae, Qodo, Command Code, Grok. Skills use the open SKILL.md format which is compatible with any AI coding agent that reads markdown instructions.