AI Agents & MCP Integration
AI agents need data — blockchain state, wallet balances, transaction histories, DeFi prices, on-chain governance data. This data has to come from somewhere reliable, fast, and uncensorable.
Pocket Network is that layer. With 60+ chains accessible through a single JSON-RPC endpoint, and support for any HTTP service as a data Source since the Shannon upgrade, Pocket is the natural data infrastructure for AI agents.
What Is MCP?
MCP (Model Context Protocol) is the open standard for connecting AI models to external tools and data sources. Build one MCP server, and any MCP-compatible AI client (Claude, GPT-4, Gemini, open-source models) can use it.
BlockchainQuery — Pocket’s MCP Server
BlockchainQuery is Pocket Network’s MCP server and Claude Desktop extension. It exposes 37 read-only blockchain query tools across 63 chains and 7 protocol families (EVM, Solana, Cosmos, Near, Sui, Radix, Tron).
What it exposes as MCP tools:
- Blockchain state queries:
get_balance,get_block,get_transaction,eth_call,eth_getLogs— standard JSON-RPC methods routed through Pocket’s decentralized supplier pool - Multi-chain queries: Query any supported chain through the same server — the AI model specifies the chain, the server routes to the correct Pocket endpoint
- DeFi data: Contract calls for DEX prices, liquidity pools, lending rates, oracle feeds
- Cross-protocol: EVM, Cosmos, Solana, Near, Sui, and Radix from a single interface
Quickstart: Claude Desktop
BlockchainQuery is available as a .mcpb extension file, following Anthropic’s standard extension installation format. Download the .mcpb file from the GitHub repository and install it using Claude’s extension installer. A listing in Anthropic’s MCP directory is pending.
Once installed, Claude can query live blockchain data: wallet balances, recent transactions, token info, governance proposals, and more — across 63 chains.
Integration with Agent Frameworks
LangChain / LangGraph
LangChain natively supports JSON-RPC tool calls. Use Pocket’s endpoint as the provider URL for any ethers.js or web3.py-based LangChain tool:
from langchain.tools import Tool
from web3 import Web3
w3 = Web3(Web3.HTTPProvider("https://eth.api.pocket.network"))
def get_balance(address: str) -> str:
balance = w3.eth.get_balance(address)
return str(w3.from_wei(balance, 'ether')) + " ETH"
tool = Tool(name="get_eth_balance", func=get_balance, description="Get ETH balance")OpenAgents / CrewAI
Both frameworks support custom tool definitions. Pocket Network endpoints work as standard HTTP tools — no library-specific wrapper needed.
x402 — Autonomous Agent Payments
x402 is the HTTP-native payment protocol that lets AI agents pay for data access autonomously — without human-in-the-loop billing. An agent sends an HTTP request; if the resource requires payment, the server returns HTTP 402; the agent pays in USDC stablecoin and retries.
PNF’s 2026 roadmap includes native x402 compatibility as a P1 initiative (Q2-26). This will allow AI agents to access Pocket’s endpoints autonomously using x402 micropayments — bypassing the need to hold or stake POKT directly.
Why Pocket Network Over Centralized Alternatives?
- No API keys or accounts — permissionless access to 60+ chains
- Decentralized — no single point of failure, no vendor lock-in
- Multi-chain — one endpoint pattern for every supported chain
- Censorship-resistant — no rate limiting by content, no geographic restrictions
- Open source — all infrastructure is auditable and forkable
Getting Started — Quick Paths
| I want to… | Do this |
|---|---|
| Query a chain from my agent | Point at https://{chain-slug}.api.pocket.network — no account needed |
| Give Claude blockchain access | Add BlockchainQuery to Claude Desktop |
| High-throughput agent access | Apply for the Foundation Partnership program or stake an Application via the application staking docs |
| Build an AI product on Pocket | Register your service as a Pocket Source — earn a share of relay rewards proportional to usage |
| Track MCP/x402 progress | Watch github.com/pokt-network/poktroll and Discord #agentic-ai |