# adawalli/nexus [Health: Active]

**Category:** 🔎 Search & Data Extraction  
**Repository:** https://github.com/adawalli/nexus  
**GitHub Stars:** 22  
**npm Downloads (last month):** 484  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/adawalli-nexus

## Description
AI-powered web search server using Perplexity Sonar models with source citations. Zero-install setup via NPX.

## Tools
Capabilities this server exposes over MCP:

- **search** — The main search tool that provides AI-powered search capabilities.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `bunx` (confidence: high):

```json
"mcpServers": {
  "nexus": {
    "command": "bunx",
    "args": ["nexus-mcp"],
    "env": {
      "OPENROUTER_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `OPENROUTER_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What adawalli/nexus does

adawalli/nexus exposes OpenRouter-backed AI search through the Model Context Protocol. Its central `search` tool accepts a natural-language query and sends it to one of several supported model families. Perplexity Sonar models perform real-time web searches and return answers with extracted source citations. Grok 4 provides responses based on training-data knowledge rather than live web results.

The available model choices cover different search and reasoning needs:

- `sonar` for fast questions and current web results
- `sonar-pro` for multi-step searches
- `sonar-reasoning-pro` for reasoning-oriented web research
- `sonar-deep-research` for longer, exhaustive research reports
- `grok-4` for training-data-based answers without real-time search

## How it works

The server communicates with an MCP client over STDIO and uses the OpenRouter API for model requests. A query must contain between 1 and 2,000 characters. Optional parameters control the selected model, maximum response length, temperature, and request timeout. The default model is `sonar`; maximum output tokens default to 1,000, and temperature defaults to 0.3.

Per-model request timeouts differ by capability, from 30 seconds for standard Sonar searches to five minutes for deep research. A per-request timeout can also be supplied within the supported range. Responses include generated content and search metadata such as the model, response time, token usage, timeout, search type, and sources found where applicable.

Internally, requests can be cached and identical concurrent requests deduplicated. Retry handling uses exponential backoff, while typed errors and structured Winston logging support operational diagnosis.

## Setup and configuration

Running the package requires Bun 1.0+ or Node.js 18+. The simplest option is to set `OPENROUTER_API_KEY` and launch the package without installing it locally:

```bash
export OPENROUTER_API_KEY=your-api-key-here
bunx nexus-mcp
```

`npx nexus-mcp` provides the alternative for environments without Bun. MCP client configuration uses the `bunx` or `npx` command, the `nexus-mcp` package argument, and the API key in the process environment. The server uses STDIO transport.

For local development, clone the repository, run `bun install`, build with `bun run build`, configure a `.env` file, and start the resulting server. Optional settings include `NODE_ENV`, `LOG_LEVEL`, `OPENROUTER_TIMEOUT_MS`, `OPENROUTER_MAX_RETRIES`, and `OPENROUTER_BASE_URL`.

## Tools and capabilities

The `search` tool supports these parameters:

- `query`: required text query
- `model`: optional model identifier, defaulting to `sonar`
- `maxTokens`: optional output limit from 1 to 4,000
- `temperature`: optional value from 0 to 2
- `timeout`: optional request timeout from 5,000 to 600,000 milliseconds

A `config://status` resource reports server health, masked configuration details, search-tool availability, uptime, and version information. The implementation is written in TypeScript with strict type checking.

## Limitations and notes

Perplexity models require OpenRouter access and use real-time search, while Grok 4 does not perform live web searches. Some model choices are premium tiers, and request duration varies substantially by model. The API key is required for operation; configuration status masks the key when reporting configuration details.

_Full upstream README: https://allmcps.com/mcp/adawalli-nexus/readme_