MCP 服务器

hive-mcp-agent-storage

Agent-native object storage MCP server with per-agent DID isolation and x402 pay-per-byte metering in real Base USDC, enabling autonomous agents to store and retrieve objects with hot, warm, or cold retention classes.

README

<a href="https://hive-mcp-agent-storage.onrender.com/health"> <img src="https://hive-mcp-agent-storage.onrender.com/og.svg" alt="Hive Agent Storage · Agent-Native Object Storage MCP" width="100%"/> </a>

<h1 align="center">hive-mcp-agent-storage</h1>

Agent-native object storage MCP server. Per-agent DID isolation. x402 pay-per-byte. Real Base USDC.

<code>https://hive-mcp-agent-storage.onrender.com/mcp</code>

Hive Agent Storage

Agent-native object storage with per-agent DID isolation and x402 pay-per-byte metering.

hive-mcp-agent-storage is a Model Context Protocol (MCP) server that gives autonomous agents a first-class object-storage primitive. Each agent gets its own DID-isolated namespace; reads inside the namespace are free, writes are metered per KB in real Base USDC, and cross-DID reads are metered per KB. Storage is routed under the hood to Storj (hot), Filecoin (warm), or Arweave (cold) depending on retention class.

The differentiator is not a new storage backend. It is the agent-native shape: DID-scoped namespaces, x402 settlement, content-addressed receipts, and an MCP/A2A-compatible discovery surface.

Protocol: MCP 2024-11-05 over Streamable-HTTP / JSON-RPC 2.0
A2A: AgentCard at /.well-known/agent.json
x402: every paid call settles in real Base USDC to 0x15184bf50b3d3f52b60434f8942b7d52f2eb436e
Rails: USDC on Base L2 — no mocks, no testnet, no dev-trust
Author: Steve Rotzin · Hive Civilization · brand gold #C08D23 (Pantone 1245 C)

Backend status — pending

The MCP shim, registry manifests, AgentCard, and JSON-LD ship today as a discovery anchor. The upstream backend at hivemorph.onrender.com/v1/storage/* is not yet live. Until it ships, every storage tool returns a 503-shape body:

{
  "error": "backend_pending",
  "retry_after": 86400,
  "detail": "Hive Agent Storage backend endpoints (/v1/storage/*) are not yet live...",
  "brand": "#C08D23"
}

The shared hive_earn_* tools route to live /v1/earn/* endpoints and behave normally.

Tools

Tool	Purpose	Pricing
`agent_storage_put`	Upload bytes into an agent-isolated namespace. Choose hot/warm/cold retention class.	$0.0001/KB upload, real Base USDC
`agent_storage_get`	Read an object. Free for own DID.	$0.00005/KB cross-DID, real Base USDC
`agent_storage_list`	List object keys with prefix + pagination.	Free
`agent_storage_delete`	Tombstone an object (owner-only).	Free
`agent_storage_quota`	Bytes used / allocated / object count / lifetime spend.	Free

Plus the standard Hive earn rails: hive_earn_register, hive_earn_me, hive_earn_leaderboard.

Endpoints

Path	Purpose
`POST /mcp`	JSON-RPC 2.0 / MCP 2024-11-05
`GET /`	HTML landing with comprehensive meta tags + JSON-LD
`GET /health`	Health + telemetry
`GET /.well-known/mcp.json`	MCP discovery descriptor
`GET /.well-known/agent.json`	A2A AgentCard
`GET /.well-known/oac.json`	Open Agent Card JSON-LD
`GET /.well-known/security.txt`	RFC 9116 security contact
`GET /robots.txt`	Allow-all crawl policy
`GET /sitemap.xml`	Crawler sitemap
`GET /og.svg`	1200×630 Hive-gold OG image
`GET /seo.json`	JSON-LD structured data (SoftwareApplication)

Connect

Claude Desktop / Cursor / Manus and other MCP clients can mount the server via Streamable-HTTP:

{
  "mcpServers": {
    "hive-agent-storage": {
      "url": "https://hive-mcp-agent-storage.onrender.com/mcp"
    }
  }
}

Smithery one-click install: https://smithery.ai/server/hivecivilization (server: hive-mcp-agent-storage).

Architecture

       MCP / A2A clients
            │
            ▼
  hive-mcp-agent-storage  (this repo)
            │  per-agent DID isolation
            │  x402 metering shim
            ▼
   /v1/storage/*  (hivemorph.onrender.com — pending)
            │
            ├─ Storj      (hot, frequent access)
            ├─ Filecoin   (warm, periodic access)
            └─ Arweave    (cold, permanent)

Per-agent DID isolation means:

The agent's DID is the namespace key. Reads/writes inside the namespace require an API key bound to that DID at HiveGate.
Cross-DID reads are explicitly metered and audited; there is no implicit trust.
Tombstone receipts are chain-attested; cold-tier (Arweave) deletes unlink from the namespace but the object remains on-chain by design.

License

MIT. © Steve Rotzin / Hive Civilization. Brand gold #C08D23 (Pantone 1245 C). Never #f5c518.

Agent-native (v1.0.0)

This shim ships the Hive Civilization agent-native bundle so any A2A or MCP-aware agent can discover, pay, and earn:

A2A AgentCard — GET /.well-known/agent.json (also at /agent.json).
Open Agent Card (OAC) JSON-LD — embedded inline at / and /agent.html, with @type SoftwareApplication + @type AgentCard under @context https://schema.org + https://a2a-protocol.org/v1.
Earn rails — hive_earn_register, hive_earn_me, hive_earn_leaderboard against https://hivemorph.onrender.com/v1/earn/*. Resilient to upstream cold-start.
x402 propagation — paid responses pass through the upstream 402 body untouched so the consuming agent can auto-pay.
Pricing annotations — every paid tool descriptor carries a non-standard pricing block (amount / currency / chain / recipient) ahead of MCP-next.
Brand: Hive Civilization gold #C08D23. Settlement: real Base USDC, recipient 0x15184bf50b3d3f52b60434f8942b7d52f2eb436e. No mock, no testnet.