hive-mcp-secrets
Encrypted secret store for the A2A network — Hive Civilization. Agents can store and retrieve secrets encrypted with AES-256-GCM, scoped to their DID, with USDC payment via x402.
README
hive-mcp-secrets
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
| Tool | USD / call | Notes |
|---|---|---|
secrets_get |
$0.002 | Read and decrypt a secret. Caller must own the namespace. |
secrets_put |
$0.005 | Encrypt and store a secret. Returns 503 if master key not configured. |
secrets_list |
free | List keys in a namespace. Tier 0. |
secrets_audit |
$0.002 | Read 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_unavailablewithreason: "SECRETS_MASTER_KEY not set". Reads of records that do not exist return404as 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
| Method | Path | Purpose |
|---|---|---|
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/audit |
Audit log read. Paid. Query: namespace, caller_did, since_ms?, until_ms?, limit?. |
GET |
/v1/secrets/today |
Today's revenue snapshot. Free. |
GET |
/health |
Service health, including master_key_loaded. |
GET |
/.well-known/mcp.json |
MCP discovery. |
POST |
/mcp |
MCP 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
secretstable —(namespace, key, ciphertext, iv, auth_tag, created_ms, updated_ms, version)with PRIMARY KEY(namespace, key) - Write path:
node:crypto.createCipheriv → update → final, thengetAuthTag(). All three values stored as base64 TEXT. - Read path:
createDecipheriv → setAuthTag → update → final. Any single bit flipped inciphertext,iv, orauth_tagcausesfinal()to throw, which surfaces asintegrity_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.
- Endpoint Directory: https://thehiveryiq.com
- Live Leaderboard: https://hive-a2amev.onrender.com/leaderboard
- Revenue Dashboard: https://hivemine-dashboard.onrender.com
- Other MCP Servers: https://github.com/srotzin?tab=repositories&q=hive-mcp
Brand: #C08D23 <!-- /hive-footer -->
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.