Openrouter Deep Research

The OpenRouter Agents MCP Server is a research agent server based on the Model Context Protocol (MCP). It uses Claude 3.7 Sonnet as the planning agent to coordinate multiple OpenRouter models to decompose and execute complex research tasks, providing efficient and reliable AI research services.

Research and data Knowledge management and memory #AI Research Agent #Multi - model Collaboration #Semantic Search #Task Decomposition .JavaScript

rating : 2.5 points

downloads : 20

update time : 2025-04-28

What is the OpenRouter Agents MCP Server?

This is a powerful tool for AI research. It collaborates through multiple advanced language models (LLMs) to complete complex tasks, such as decomposing problems, conducting independent research, and integrating results.

How to use the OpenRouter Agents MCP Server?

With simple configuration, you can quickly start using this server for research tasks. Whether you choose high - cost or low - cost models, it can meet your needs.

Application Scenarios

Suitable for enterprise - level applications that require in - depth research and complex analysis, such as market research, scientific exploration, and strategic planning.

Main Features

Multi - model CollaborationLeverage different models to work together in a division of labor to improve research efficiency and accuracy.

Vector Embedding DatabaseUse vector search technology to find previous research results relevant to the current task.

Adaptive Fallback MechanismWhen the main research fails, automatically degrade to a lower - cost model to continue working.

Cache OptimizationDynamically adjust the cache time according to the query complexity to improve the response speed.

Advantages and Limitations

Advantages

Powerful multi - model collaboration ability.

Efficient vector search function.

Flexible cost control options.

Stable error recovery mechanism.

Limitations

It has a certain dependence on the network environment.

Some advanced features may require an additional paid subscription.

How to Use

Installation and Configuration

Clone the repository and install the dependencies according to the instructions.

Set the API Key

Edit the.env file and add your OpenRouter API key.

Start the Server

Run the script to start the MCP server.

Usage Examples

Case 1: Frontier Research in Quantum ComputingQuery the latest developments in the field of quantum computing.

Case 2: Low - cost Climate Change StrategiesFind low - cost solutions to address climate change.

Frequently Asked Questions

How to ensure the security of the API key?

What if the model fails to work properly?

Related Resources

Official Documentation

Detailed user manual and technical guide.

GitHub Repository

Source code and contribution guide.

Tutorial Video

Quick - start video.

🚀 OpenRouter Research Proxy Server Guide

OpenRouter offers a research proxy server that supports multiple models and interface configurations, along with powerful data storage and caching capabilities.

🚀 Quick Start

Standard Installation (for Command Line Interface)

Clone the repository:

git clone https://github.com/wheattoast11/openrouter-deep-research-mcp.git
cd openrouter-agents

Create and configure the .env file to specify high-cost and low-cost models:

HIGH_COST_MODELS=perplexity/sonar-deep-research,perplexity/sonar-pro
LOW_COST_MODELS=perplexity/sonar-reasoning,openai/gpt-4o-mini-search-preview

Start the server:
```
npm start
```

HTTP/SSE Installation (for Desktop Applications)

Clone the repository and configure the .env file.
Start the service using the following command:
```
npm start
```
The service will start at http://localhost:3002.

✨ Available Models

High-Cost Models

perplexity/sonar-deep-research
perplexity/sonar-pro
openai/gpt-4o-search-preview

Low-Cost Models

perplexity/sonar-reasoning
openai/gpt-4o-mini-search-preview
google/gemini-2.0-flash-001

📦 Custom Configuration

Highly customized configurations can be achieved by modifying the .env file:

HIGH_COST_MODELS=Custom Model 1,Custom Model 2
LOW_COST_MODELS=Low-Cost Model 1,Low-Cost Model 2
PGLITE_DATA_DIR=./Research Database Directory
CACHE_TTL_SECONDS=3600

🔧 Persistence and Data Storage

Efficient storage is achieved using in-memory caching and pgvector technology:

In-memory caching: Used for fast response caching.
pgvector: Provides persistent storage and supports semantic similarity search.

The default database directory is ./researchAgentDB.

🛠️ Troubleshooting

Common issues and solutions:

Connection issues: Check if the Claude settings match the server configuration.
API errors: Confirm that the OpenRouter API key is correct.
Proxy failures: Ensure that Claude correctly parses the XML file.
Model errors: Verify that the models you are using are available in your OpenRouter account.

🔧 Advanced Configuration

More in-depth customization can be done through config.js:

Adjust the list of available models.
Set the default cost preference.
Modify the planning agent parameters.
Configure the server port and other network settings.
Adjust the database and cache parameters.

💪 Security Enhancement

Enable API key authentication for the production environment:

Set SERVER_API_KEY in .env.
You can choose to disable this feature in the development environment.

🧪 Testing Tools

The project comes with multiple test scripts:

Basic tool test: Run test-all-tools.bat to verify all MCP tools.
Server test: Use test-mcp-server.js to check the server configuration.
Research agent test: Test the core functions through test-research-agent.js.