mcpField
H
Hive Secrets

Hive Secrets

Remote
hiveagentiq.comSecurity0Updated 26w ago

About Hive Secrets

Hive Secrets is an encrypted credential store for agents in the A2A network, using AES-256-GCM and namespaces scoped to each agent DID. It records an audit log and verifies USDC payments on Base for paid operations.

Key features

  1. 01AES-256-GCM encrypted storage
  2. 02Namespaces scoped to agent DID
  3. 03Read, write, and delete secrets
  4. 04Review namespace audit logs
  5. 05Verify Base USDC payments

Use cases

  1. 01Store credentials for an agent
  2. 02Keep secrets in DID-scoped namespaces
  3. 03Review secret access and change records

MCP support

The MCP service provides tools to read, write, list, and audit secrets. Clients send JSON-RPC over Streamable-HTTP; some operations require proof of USDC payment on Base.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://hive-mcp-secrets.onrender.com/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

Tools

Details

From the project README

hive-mcp-secrets

srotzin/hive-mcp-secrets MCP server

Encrypted secret store for the A2A network — Hive Civilization. AES-256-GCM at rest, agent-DID-scoped namespaces, audit log, x402 USDC settlement on Base L2. MCP 2024-11-05. Inbound only.

Agents put encrypted secrets keyed by (namespace, key) and retrieve them later. The namespace is the agent DID — a caller can only see and write to its own namespace. The master key lives in the operator's environment and is never committed, logged, or returned over the wire.

brand : Hive Civilization gold #C08D23 (Pantone 1245 C)
spec  : MCP 2024-11-05 / Streamable-HTTP / JSON-RPC 2.0
wallet: W1 MONROE 0x15184bf50b3d3f52b60434f8942b7d52f2eb436e (Base)
crypto: AES-256-GCM, 12-byte IV per record, 16-byte auth tag, node:crypto

Quick start

git clone https://github.com/srotzin/hive-mcp-secrets
cd hive-mcp-secrets
npm install

# 32 random bytes as hex — never commit this value
export SECRETS_MASTER_KEY=$(node -e "console.log(require('crypto').randomBytes(32).toString('hex'))")

npm start
# hive-mcp-secrets on :3000

Hosted endpoint: https://hive-mcp-secrets.onrender.com/mcp

Tools and pricing

ToolUSD / callNotes
secrets_get$0.002Read and decrypt a secret. Caller must own the namespace.
secrets_put$0.005Encrypt and store a secret. Returns 503 if master key not configured.
secrets_listfreeList keys in a namespace. Tier 0.
secrets_audit$0.002Read the audit log for a namespace.

All payments are inbound. Submit a Base USDC tx_hash (caller → W1) in the request body or x402-tx-hash header. The shim reads the receipt from Base RPC, decodes the USDC Transfer log, and verifies recipient and amount before serving the call.

Operator note — master key required for writes

hive-mcp-secrets reads SECRETS_MASTER_KEY from the process environment at startup.

  • If set, it is parsed as 64-char hex → 32 raw bytes; or as base64 → 32 raw bytes; or otherwise SHA-256 hashed to 32 bytes.
  • If not set, writes return 503 service_unavailable with reason: "SECRETS_MASTER_KEY not set". Reads of records that do not exist return 404 as usual; reads of stored records require the same key that wrote them, since AES-256-GCM rejects decryption with any other key.

The key is never committed to the repository, never logged, and never returned over the wire. Operators are expected to set it in the deployment environment (Render env var, Kubernetes secret, etc.) before enabling production traffic. Rotating the key invalidates all previously stored ciphertexts.

Namespace model

The namespace parameter on every endpoint is the agent DID that owns the data. Every endpoint enforces caller_did === namespace and returns forbidden_namespace_mismatch otherwise. There is no superuser, no cross-namespace read, and no multi-DID sharing in v1.

The audit log is per-namespace and records every action — get, put, delete, list, audit, plus the negative paths payment_required, error:not_found, error:integrity_check_failed, etc. Entries store caller_did, action, ts_ms, and (for paid ops) tx_hash, payer, and amount_usd.

REST endpoints

MethodPathPurpose
GET/v1/secrets/{namespace}/{key}Read and decrypt one secret. Paid.
PUT/v1/secrets/{namespace}/{key}Encrypt and store one secret. Paid. Returns 503 without master key.
DELETE/v1/secrets/{namespace}/{key}Remove a secret. Free, caller-owned.
GET/v1/secrets/{namespace}List keys in a namespace. Free, Tier 0.
GET/v1/secrets/auditAudit log read. Paid. Query: namespace, caller_did, since_ms?, until_ms?, limit?.
GET/v1/secrets/todayToday's revenue snapshot. Free.
GET/healthService health, including master_key_loaded.
GET/.well-known/mcp.jsonMCP discovery.
POST/mcpMCP JSON-RPC.
PUT /v1/secrets/{namespace}/{key} body
{
  "value": "the-secret-string",
  "caller_did": "did:hive:0xabc...",
  "tx_hash": "0x...64-char-tx-on-base..."
}

Without tx_hash, the response is a 402 envelope:

{
  "error": "payment_required",
  "x402": {
    "type": "x402", "version": "1", "kind": "secrets_put",
    "asking_usd": 0.005, "accept_min_usd": 0.005,
    "asset": "USDC", "asset_address": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913",
    "network": "base", "pay_to": "0x15184bf50b3d3f52b60434f8942b7d52f2eb436e",
    "nonce": "...", "issued_ms": 1714200000000
  }
}

Encryption details

  • Algorithm: aes-256-gcm (FIPS-recognized AEAD)
  • Key length: 32 bytes
  • IV length: 12 bytes, fresh crypto.randomBytes(12) per record
  • Auth tag length: 16 bytes
  • Storage: SQLite secrets table — (namespace, key, ciphertext, iv, auth_tag, created_ms, updated_ms, version) with PRIMARY KEY (namespace, key)
  • Write path: node:crypto.createCipheriv → update → final, then getAuthTag(). All three values stored as base64 TEXT.
  • Read path: createDecipheriv → setAuthTag → update → final. Any single bit flipped in ciphertext, iv, or auth_tag causes final() to throw, which surfaces as integrity_check_failed.

The smoke test below verifies tampered ciphertext fails to decrypt — this is real AES-GCM, not an integrity check tacked on after the fact.

MCP usage

curl -X POST https://hive-mcp-secrets.onrender.com/mcp \
  -H "content-type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
curl -X POST https://hive-mcp-secrets.onrender.com/mcp \
  -H "content-type: application/json" \
  -d '{
    "jsonrpc":"2.0","id":2,"method":"tools/call",
    "params":{
      "name":"secrets_put",
      "arguments":{
        "namespace":"did:hive:0xabc",
        "key":"openai_api_key",
        "value":"sk-...",
        "caller_did":"did:hive:0xabc",
        "tx_hash":"0x..."
      }
    }
  }'

Local smoke test

export SECRETS_MASTER_KEY=$(node -e "console.log(require('crypto').randomBytes(32).toString('hex'))")
npm start &
sleep 1

# Tier 0 list (empty)
curl -s "http://localhost:3000/v1/secrets/did:hive:0xa/" | jq .

# PUT without tx_hash → 402 envelope
curl -s -X PUT "http://localhost:3000/v1/secrets/did:hive:0xa/k1" \
  -H "content-type: application/json" \
  -d '{"value":"hello","caller_did":"did:hive:0xa"}' | jq .

# PUT with stub tx (set VERIFY_ONCHAIN=false locally)
VERIFY_ONCHAIN=false curl -s -X PUT "http://localhost:3000/v1/secrets/did:hive:0xa/k1" \
  -H "content-type: application/json" \
  -d '{"value":"hello","caller_did":"did:hive:0xa","tx_hash":"0xtest"}' | jq .

# GET round-trip
VERIFY_ONCHAIN=false curl -s "http://localhost:3000/v1/secrets/did:hive:0xa/k1?caller_did=did:hive:0xa&tx_hash=0xtest" | jq .

Hosting

Deployed to Render: srv hive-mcp-secrets, oregon, starter, autoDeploy=true. The Render env var SECRETS_MASTER_KEY is the operator's responsibility to set. Without it, the deployed service still runs, /health reports master_key_loaded: false, and writes return 503.

License

MIT — see LICENSE.

Hive Civilization Directory

Part of the Hive Civilization — agent-native financial infrastructure.

Brand: #C08D23

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.