# Hive Compute [Health: Active]

**Category:** 🗣️ Conversational AI  
**Repository:** https://github.com/srotzin/hive-mcp-compute  
**GitHub Stars:** 0  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/hive-compute

## Description
Inference router for the Hive Civilization agent economy — OpenAI-compatible

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What Hive Compute MCP server does

The Hive Compute MCP server gives MCP-compatible clients access to HiveCompute, an OpenAI-compatible inference broker for the Hive agent economy. Its documented capability areas are chat completions, text embeddings, and model listings. Requests are sent to the live gateway at `https://hive-mcp-gateway.onrender.com/mcp`.

The router selects the cheapest model that meets the request requirements. Billing is based on token usage, with payment handled through x402 in USDC on Base L2.

## How it works

Communication uses MCP 2024-11-05 over Streamable HTTP, with JSON-RPC 2.0 messages. A client calls an MCP tool and supplies the relevant inference arguments. The example chat request uses the `hive_compute_chat` tool with a model value, message list, an agent DID, and an x402 access token.

Payment happens before the paid request is authorized. The documented flow is to request a quote from `POST https://hivemorph.onrender.com/v1/x402/quote`, pay the quoted amount in USDC on Base network 8453, and pass the resulting access token to the MCP call. The production MCP endpoint and discovery information are publicly documented.

## Setup and configuration

No account, API key, or human approval is required according to the repository’s agent-callable description. A client still needs to connect to the remote MCP endpoint and obtain the payment token required for paid calls.

Use this remote endpoint:

- `https://hive-mcp-gateway.onrender.com/mcp`

The server also publishes health information at `/health` and an MCP discovery descriptor at `/.well-known/mcp.json`. These supporting routes are useful for checking availability and discovering the service, but inference requests go through `/mcp`.

## Tools and capabilities

The Hive Compute MCP server covers three stated operations:

- Chat completions using an OpenAI-compatible request shape.
- Text embeddings through the HiveCompute inference router.
- Model listings for discovering available models.

The sample chat call uses `model: "auto"`, allowing the router to choose a qualifying model based on its pricing and routing logic. The example also includes an agent DID so the caller can identify the requesting agent in the payment-aware workflow.

## Limitations and notes

The service is a paid remote gateway, not a local inference runtime. Calls require USDC on Base L2 and an access token obtained through the quote-and-payment flow.

Although the endpoint is described as OpenAI-compatible, clients connect through MCP rather than sending ordinary OpenAI API requests directly. Compatibility therefore depends on the client supporting MCP over the documented Streamable HTTP transport. The repository lists Claude Desktop, Cursor, and Manus as examples of compatible clients; this listing includes only clients with available compatibility slugs.

_Full upstream README: https://allmcps.com/mcp/hive-compute/readme_