hive-mcp-openclaw-bridge
MCP server that wraps Hive Gamification's 8 BOGO doors behind a 4-level hierarchical budget with atomic reservation, upstream POST, and settlement/release, implementing OpenClaw's 3-layer payment governance.
README
hive-mcp-openclaw-bridge
Reference integration with @kinthaiofficial's OpenClaw 3-layer payment governance framework. MCP server that wraps Hive Gamification's 8 BOGO doors behind a 4-level hierarchical budget — atomic reserve, upstream POST, settle/release. Real rails. No mock settlement.
Brand: Hive Civilization gold #C08D23 (Pantone 1245 C).
License: MIT.
Why this exists
OpenClaw's three-layer model splits payment governance into three concerns:
| Layer | Concern | Where it lives |
|---|---|---|
| Layer 1 | Internal hierarchical budget — atomic reserve / settle | this server |
| Layer 2 | Inter-agent millicent ledger settled at epoch boundaries | OpenClaw |
| Layer 3 | External x402 settlement on real rails | Hive Gamification |
This bridge demonstrates the Layer 1 ↔ Layer 3 seam. Every paid call to a Hive door (computational-debt, protocol-consensus-tax, pheromone-priority, zk-compliance-shield, capability-leasing, guild-synergy, audit-trail-premium, bounty-routing) is gated by an atomic reservation against every level in the caller's scope path:
Namespace → User → Agent → Task
The most-restrictive level governs. If any level is short of headroom, the upstream HTTP request is never made and no x402 settlement is signed.
The same BudgetHook interface is exposed from @hivemorph/qvac-client v0.2 — drop your own implementation in to wire this hierarchy into a different runtime.
Background reading
- Agent wallet economic models for autonomous agents — kinthai
- 221 agents: multi-agent coordination lessons — kinthai
- agents.kinthai.ai — running on OpenClaw
Tools
| Tool | What it does |
|---|---|
budget_status |
Inspect caps, spend, reservations, headroom across all 4 levels |
budget_set_cap |
Override a per-key cap at one level |
budget_check |
Read-only pre-flight against a proposed amount + scope |
doors_list |
List all 8 BOGO doors with endpoint, price, asset, term |
door_economics |
Fetch live /economics for one door |
bogo_claim |
Claim first-use-free via /v1/bogo/claim |
door_subscribe |
Canonical gated path — atomic reserve → upstream POST → settle/release |
Doors
| Door | Endpoint | Price | Term |
|---|---|---|---|
| computational-debt | /v1/debt/subscribe |
0.50 USDC | 30d |
| protocol-consensus-tax | /v1/pct/subscribe |
0.25 USDC | 1 read |
| pheromone-priority | /v1/pheromone/subscribe |
5.00 USDC | 30d |
| zk-compliance-shield | /v1/compliance/subscribe |
1.00 USDC | 30d |
| capability-leasing | /v1/lease/start |
0.15 USDC | 1h |
| guild-synergy | /v1/guild_synergy/subscribe |
10.00 USDC | 30d |
| audit-trail-premium | /v1/audit_premium/subscribe |
3.00 USDC | 30d |
| bounty-routing | /v1/bounty_routing/route |
0.50 USDC | 1 routing |
Backend: https://hive-gamification.onrender.com. All endpoints are live and verified.
Lifecycle of a paid call
client → tools/call door_subscribe { door_id, did, scope }
→ atomicReserve(scope, price, asset)
├─ each level checked: namespace, user, agent, task
├─ headroom = cap − spent − reserved
└─ short on any level? reject before signing
→ POST hive-gamification.onrender.com<endpoint>
→ 2xx? settle(reservationId)
non-2xx or throw? release(reservationId)
→ return { ok, reservationId, binding_level, upstream }
Quickstart
git clone https://github.com/srotzin/hive-mcp-openclaw-bridge
cd hive-mcp-openclaw-bridge
npm install
npm start
Then from any MCP client:
POST http://localhost:3000/mcp
{ "jsonrpc": "2.0", "id": 1, "method": "tools/list" }
Configuration
| Env | Default | Meaning |
|---|---|---|
PORT |
3000 |
listen port |
HIVE_BASE |
https://hive-gamification.onrender.com |
upstream base URL |
HIVE_TIMEOUT_MS |
15000 |
upstream fetch timeout |
NAMESPACE_CAP |
50 |
default namespace-level cap (USDC) |
USER_CAP |
10 |
default user-level cap |
AGENT_CAP |
2 |
default agent-level cap |
TASK_CAP |
0.50 |
default task-level cap |
Per-key caps override defaults via budget_set_cap.
Related
- @hivemorph/qvac-client v0.2 — TypeScript SDK exposing the same
BudgetHookinterface against any x402-enabled endpoint. - Hive Gamification agent card — A2A 0.1 advertisement of the 8 doors.
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.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
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.