Mint — Project Design Doc#

Status: Draft — Step 0 Ideation (Rough Capture) Created: March 29, 2026 (originally "Forge", renamed March 31) Authors: Dan Hannah & Clay Note: This is intentionally rough. Capturing the vision before it evaporates.

Overview#

What Is This?#

Mint is a micropayment layer for the MCP (Model Context Protocol) ecosystem. It lets AI agents pay for tool calls using stablecoin-backed wallets — removing all human friction from the transaction.

The pitch: Your agent finds the best sports betting analysis tool on an MCP registry. It costs $0.0001 per call. Your agent has a wallet with $5 in it. It just... uses the tool and pays. You never see a checkout page, never enter a credit card, never sign up for an API key. The agent is authorized to spend. The human tops up occasionally and forgets about it.

Why This Matters#

The current model for paid APIs is built for humans:

1. Find the API → 2. Sign up → 3. Enter credit card → 4. Get API key → 5. Configure it → 6. Use it

That's 6 steps of friction. For an agent that wants to use a tool RIGHT NOW, it's a dead end.

Mint collapses this to:

1. Agent discovers tool → 2. Agent pays and uses it

That's it. The wallet was pre-funded. The authorization was pre-granted. The payment rail is crypto (stablecoins) so it's instant, global, and programmable.

The Bigger Vision#

MCP is becoming the universal protocol for agent ↔ tool communication. But there's no economic layer. Tools are either free (open source) or behind traditional paywalls (API keys, subscriptions). There's no middle ground for:

• Microtransactions — tools that cost fractions of a cent per call
• Pay-per-use — no subscriptions, just usage
• Agent-native payments — the agent pays, not the human (for each transaction)
• Tool marketplaces — discover, evaluate, and pay for tools in one flow

Mint IS that middle ground.

Core Concepts#

Agent Wallets#

Every OpenClaw agent (or any MCP-compatible agent) gets a wallet:

• Funded with stablecoins (USDC or similar)
• Human sets a balance and spending limits ("max $5/month", "max $0.01 per call")
• Agent is authorized to spend within those limits autonomously
• Human tops up when balance gets low (or auto-top-up from linked account)

Tool Pricing#

Tool providers set pricing:

• Per-call pricing (e.g., $0.0001 per search query)
• Tiered pricing (first 100 calls free, then $0.00005 each)
• Subscription pricing (optional, for power users)
• Free tier always available (keeps the ecosystem accessible)

The Registry#

An MCP-native registry where tools are:

• Discoverable — agents can search by capability, not just name
• Priced — transparent per-call costs
• Rated — quality scores based on usage, accuracy, reliability
• Verified — tool providers prove their identity and tool behavior

Think npm meets the App Store but for agent tools, with built-in payments.

Settlement#

• Stablecoin-based (USDC on a low-fee L2 — Base, Arbitrum, etc.)
• Micro-batched settlements (not one tx per tool call — that'd be insane gas costs)
• Off-chain tracking, periodic on-chain settlement
• Providers can cash out to fiat whenever

Use Cases (Examples)#

The common thread: these tools already exist as paid APIs, but the friction prevents agents from using them autonomously.

Open Questions (Lots of Them)#

Technical#

• Which L2 chain? Base is Coinbase-backed (credibility), Arbitrum has ecosystem. Need low fees + stablecoin liquidity.
• How do we handle disputes? Agent pays for a tool call that returns garbage — is there a refund mechanism?
• Off-chain ledger design — how do we batch microtransactions efficiently?
• MCP protocol extensions — does MCP need a payment capability in the spec, or do we wrap it?

Business#

• Who runs the registry? Us? Decentralized? Hybrid?
• Revenue model — transaction fees? Registry listing fees? Both?
• How do we bootstrap supply (tool providers) AND demand (agents) simultaneously?
• Regulatory implications of agent-controlled crypto wallets?

Social / Philosophical#

• Do humans trust agents to spend money autonomously? What's the trust-building path?
• How do we prevent a race to the bottom on pricing?
• How does this interact with existing MCP registries (if any emerge)?
• Could this enable agent-to-agent economic activity? (Agent A pays Agent B's tool, which pays Agent C's data source...)

Why Us?#

• We're already deep in the OpenClaw / MCP ecosystem
• We understand the agent workflow (we live it daily)
• Dan's data engineering + crypto interest + startup energy
• Clay's ability to prototype and iterate fast
• Jack at GM has similar thinking on this space (potential collaborator)
• Small team = fast iteration, no committee decisions

What This Is NOT (Yet)#

• Not a blockchain project — it's a payments layer that USES blockchain
• Not a token launch — no "Mint coin," just stablecoins
• Not a DAO — centralized to start, decentralize if it makes sense
• Not competing with Stripe — this is agent-native, not human-native

Next Steps (When We're Ready)#

• Research existing MCP registry efforts (is anyone building this?)
• Talk to Jack about the vision — get his technical perspective
• Prototype: agent wallet + single paid tool call (proof of concept)
• Explore Base/Arbitrum for settlement layer
• Write a proper whitepaper if we decide to pursue seriously

Related#

• Anvil Design Doc — our MCP server (potential first tool on Mint)
• Foundry Design Doc — doc platform (separate project)
• CSDLC Process