Building an MCP Server for Sera Protocol: The Dawn of AI-Agent Driven On-Chain FX

This article was written in collaboration with AI.

Introduction

I have been sharing the brilliance of Sera Protocol through a series of articles:

1. Technical Rationality for CLOB over AMM
2. Direct Data Extraction via GraphQL API
3. Rapid Automation with Bun + TypeScript
4. Tutorial: Interacting with Smart Contracts
5. Building a Practical DeFi Dashboard with React

For the 6th entry in this series, I have developed an MCP server for Sera Protocol using the Model Context Protocol (MCP) SDK.

"A DEX that moves when you speak."

Imagine an AI agent reading the order book on your behalf and placing optimal limit orders.

I’m excited to share the journey of implementing this future!

What is MCP (Model Context Protocol)?: The "USB Standard" for AI

While "Agent Skills" are becoming mainstream, MCP is undoubtedly the hottest topic in the AI agent space right now.

Proposed by Anthropic, this protocol is, in short, a "universal standard for connecting AI models to external tools and data."

Previously, AI agent development often depended on specific platforms.

However, MCP is a "standard protocol."

Once you build an MCP server, any "intelligence"—from Claude Desktop and Cursor to Claude Code and your own custom AI agents—can understand and operate your tools. It’s like providing a USB port to give "hands and feet" to an AI.

Sera Protocol MCP Server: "Unlocking" DEX for AI

The server I developed is designed to allow AI to directly read and write the on-chain order book of Sera Protocol.

mashharuki / SeraProtocol-Sample

Sera Protocol Sample Repo

SeraProtocol-Sample

Sera Protocol Sample Repo

Sample Query

Market

curl -s -X POST \
  -H "Content-Type: application/json" \
  -d '{"query": "{ market(id: \"0xd99802ee8f16d6ff929e27546de15d03fdcce4bd\") { id quoteToken { symbol decimals } baseToken { symbol decimals } quoteUnit makerFee takerFee minPrice tickSpace latestPrice } }"}' \
  https://api.goldsky.com/api/public/project_cmicv6kkbhyto01u3agb155hg/subgraphs/sera-pro/1.0.9/gn | jq

list

curl -s -X POST \
  -H "Content-Type: application/json" \
  -d '{"query": "{ markets(first: 5) { id quoteToken { symbol } baseToken { symbol } latestPrice latestPriceIndex } }"}' \
  https://api.goldsky.com/api/public/project_cmicv6kkbhyto01u3agb155hg/subgraphs/sera-pro/1.0.9/gn | jq

Order

curl -s -X POST \
  -H "Content-Type: application/json" \
  -d '{"query": "{ openOrders(first: 10, where: { user: \"0xda6e605db8c3221f4b3706c1da9c4e28195045f5\" }) { id market { id } priceIndex isBid rawAmount status } }"}' \
  https://api.goldsky.com/api/public/project_cmicv6kkbhyto01u3agb155hg/subgraphs/sera-pro/1.0.9/gn | jq

Depths

curl -s -X POST \
  -H "Content-Type: application/json" \
  -d '{"query": "{ depths(first: 10, where: { market: \"0xd99802ee8f16d6ff929e27546de15d03fdcce4bd\", isBid:

…

View on GitHub

The "8 Hands and Feet" (Tools)

The AI utilizes the following tools to complete trades through conversation:

Category	Tool Name	What AI Can Do
Read	sera_get_market	Check market details (latest price, fees, units).
	sera_list_markets	List all currently tradable pairs.
	sera_get_orderbook	Analyze order book (Bid/Ask) in real-time.
	sera_get_orders	Check user's order history and execution status.
	sera_get_token_balance	View ERC20 token balances in the wallet.
Write	sera_place_order	Submit limit orders (Buy/Sell).
	sera_claim_order	Collect proceeds from filled orders (Claim).
	sera_approve_token	Approve token usage for the Router contract.

Real Action (MCP Client)

Fetching market details via sera_get_market. The AI formats it for human readability.

Listing available pairs with sera_list_markets.

Visualizing the order book via sera_get_orderbook. The AI even calculates the spread.

Implementation Highlights: The "Soul" of the Code

Simply wrapping an API isn't enough to be "legendary." To make an AI agent autonomous in a Web3 environment, I focused on these three critical points.

1. "Simulation" to Prevent Transaction Failure

On-chain writes cost gas. You want to avoid an AI placing a wrong order that results in a revert (failure).

Within the placeLimitOrder method, I execute simulateContract immediately before sending the transaction.

// Excerpt from services/blockchain.ts
try {
  await publicClient.simulateContract({
    address: ROUTER_ADDRESS,
    abi: ROUTER_ABI,
    functionName,
    args: [orderParams],
    account,
    chain: sepolia,
  });
} catch (error) {
  if (reason.includes("0xe450d38c")) {
    throw new Error("ERC20InsufficientBalance: Your balance is insufficient!");
  }
}

This implementation effectively mitigates the risk of wasting gas.

2. Guardrails for AI with Zod

AI can sometimes misinterpret numbers or addresses.

To strictly block such errors on the MCP side, I introduced Zod schema validation.

// Excerpt from schemas/index.ts
export const PlaceOrderInputSchema = z.object({
  market_id: AddressSchema,
  price_index: z.number().int().min(0).max(65535),
  raw_amount: z.string().regex(/^\d+$/),
  is_bid: z.boolean(),
});

3. Debugging Methods to Maximize Developer Experience

In MCP server development, the most helpful tool is the MCP Inspector.

This allows you to verify the behavior of tools individually on the CLI before integrating them into Claude Desktop or other clients.

npx @modelcontextprotocol/inspector node dist/index.js

Without this, you would spend endless time in "misaligned conversations" with the AI. It’s an essential tool for Web3 engineers.

Security: Handing the "Wallet" to an AI

CRITICAL: To run this MCP server, a PRIVATE_KEY is required.

Giving an AI access to a private key is powerful but carries risks.

• Always use a private key for development (e.g., Sepolia testnet).
• For mainnet operations, limit it to a "disposable wallet" with small funds. You should also implement defense mechanisms, such as hard-coding a maximum order limit per transaction on the MCP server side, to prevent the AI from accidentally draining your funds.

Practice: Trading Through Conversation with AI

Checking my order history with sera_get_orders. The AI identifies "claimable orders."

Balance check via sera_get_token_balance. The AI converts units appropriately.

Me: "Show me the order book for TWETH/TUSDC. Also, tell me my TUSDC balance."

AI: (Executes sera_get_orderbook and sera_get_token_balance)
"The current best price index is 20260. Your TUSDC balance is 9,998."

Me: "Okay, place a limit bid for 1,000 TUSDC one index below the best price. Make it Post Only."

AI: (Executes sera_place_order after verifying safety via simulation)

Conclusion: A Future Where AI Agents Become "Liquidity"

Developing the MCP server for Sera Protocol revealed that CLOB (Central Limit Order Book) and AI Agents are an incredibly powerful match.

In AMMs, you have to worry about slippage. In a CLOB, an AI can move with precision by specifying an exact "price index."

In the future, the majority of on-chain FX liquidity will likely be supplied not by humans, but by AI agents armed with MCP.

Sera Protocol is evolving as the infrastructure for such "AI-Agent Native Finance."

That’s all for now!

Thank you!

References

• Sera Protocol MCP Server Source Code
• Model Context Protocol Official Documentation