# juspay/neurolink [Health: Active]

**Category:** 🔗 Aggregators  
**Repository:** https://github.com/juspay/neurolink  
**GitHub Stars:** 157  
**npm Downloads (last month):** 22621  
**Views:** 5  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/juspay-neurolink

## Description
Making enterprise AI infrastructure universally accessible. Edge-first platform unifying 12 providers and 100+ models with multi-agent orchestration, HITL workflows, guardrails middleware, and context summarization.

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

```json
"mcpServers": {
  "neurolink": {
    "command": "npx",
    "args": ["-y","@juspay/neurolink"]
  }
}
```

## Documentation

## What juspay/neurolink does

juspay/neurolink exposes NeuroLink’s TypeScript-first inference layer through MCP. It gives an MCP-connected application a common interface for sending requests to multiple hosted providers, aggregator services, and local model runtimes instead of coupling the application to one vendor’s API.

The inference surface has three distinct modes. `generate` returns text, `stream` delivers incremental output, and `decide` produces a typed judgment such as a boolean, choice, or score. Decision requests can include application state and named questions, making them suitable for routing, gating, or approval logic where a calibrated answer is more useful than generated prose.

## How it works

NeuroLink places providers behind a shared request layer. A provider can be selected per request, allowing an application to move between services such as OpenAI, Anthropic, Google, AWS Bedrock, Azure, Mistral, Ollama, LM Studio, and llama.cpp. LiteLLM and OpenRouter provide access to larger model catalogs through aggregator endpoints.

A model registry supplies metadata for supported models and aliases, including context-window information used during request handling. Provider capabilities declare which inference kinds they support rather than treating every provider as interchangeable.

The MCP connection layer supports stdio, HTTP, SSE, and WebSocket transports. NeuroLink also supports MCP-compliant tool servers, so an application can combine its provider calls with external MCP tools. Its built-in optimization behavior includes model routing, context compaction, and tool selection; these operations are designed to fail open when their optional configuration is unavailable.

## Setup and configuration

The project is distributed as the `@juspay/neurolink` npm package and provides both a TypeScript SDK and a CLI. The documented entry points are the NeuroLink quick-start guide and the package’s npm page. Provider credentials are configured according to the services selected for a deployment; local runtimes can be used without relying on a hosted provider.

The MCP layer can be used with clients that support standard MCP transports. Choose the transport supported by the client and deployment environment, then configure the selected provider credentials and model settings for the requests the client will issue.

## Tools and capabilities

- Generate text through a selected provider.
- Stream incremental model output.
- Request typed boolean, choice, or score decisions.
- Route requests across hosted, aggregator, and local providers.
- Use model metadata, aliases, and context-window checks.
- Connect to MCP tools over stdio, HTTP, SSE, or WebSocket.
- Apply optional routing, context compaction, and tool-selection behavior.
- Use Redis memory and multi-provider failover in supported deployments.

## Limitations and notes

Provider capabilities differ: not every provider necessarily supports all three inference modes. Hosted providers require their respective credentials, while local runtimes require a running Ollama, LM Studio, or llama.cpp setup. Decision inference uses dedicated decision providers or models rather than treating ordinary text generation as an equivalent decision API.

NeuroLink is MIT licensed. Its primary development interfaces are TypeScript and the professional CLI, while MCP provides the connection surface for compatible agent clients.

_Full upstream README: https://allmcps.com/mcp/juspay-neurolink/readme_