AI & MCP Integration | OpenScan.AI - Your Guide to the AI-Powered Blockchain Explorer

Give AI assistants structured access to blockchain data.

OpenScan.AI is AI-powered by design: the explorer UI uses AI to summarize transactions and explain contract interactions, and the same structured data is exposed so you can wire AI assistants — Claude, GPT, or your own agents — directly into explorer data.

AI features in the explorer

Transaction summaries — plain-language explanations of what a transaction did: which methods ran, what value moved, which tokens changed hands.
Contract explanations — summaries of verified contract behavior from source and ABI.
Natural-language search — the search bar accepts questions and entity descriptions, not just hashes.

These features work on any instance, including self-hosted ones.

MCP server

OpenScan.AI provides a Model Context Protocol (MCP) server that exposes explorer data as tools an AI assistant can call. Tools map to REST API endpoints — looking up addresses, fetching transactions, reading verified contract source, and searching entities.

Configuration

Add the server to your MCP client (Claude Desktop, Cursor, or another MCP-capable tool):

{
  "mcpServers": {
    "openscan": {
      "command": "npx",
      "args": ["-y", "@openscanai/mcp-server"],
      "env": {
        "OPENSCAN_API_URL": "https://xdcscan.io/api/v2",
        "OPENSCAN_API_KEY": "osk_live_..."
      }
    }
  }
}

OPENSCAN_API_KEY is optional but recommended — it raises rate limits (see API Keys & Rate Limits). For a self-hosted instance, point OPENSCAN_API_URL at your own deployment.

Example tools

Tool	Backs onto
`get_address`	`GET /api/v2/addresses/{hash}`
`get_transaction`	`GET /api/v2/transactions/{hash}`
`get_contract_source`	`GET /api/v2/smart-contracts/{hash}`
`search`	`GET /api/v2/search`
`get_token_holders`	`GET /api/v2/tokens/{hash}/holders`

With the server connected, prompts like “summarize the last five transactions of this address and flag anything unusual” or “explain what this contract’s withdraw function does” resolve through live explorer data instead of the model’s guesses.

Rolling your own agent

The REST API is all an agent needs — MCP is a convenience layer. For custom agents:

Give the agent the interactive API docs (xdcscan.io/api-docs) as context.
Start with search for entity resolution, then fetch the specific resource.
Use the decoded_input and log decoding on transaction endpoints so the agent works with structured data rather than raw calldata.

See SDK Examples for request patterns to seed agent tooling.

Rate limits and keys

AI agents are bursty — a single user question can fan out into dozens of calls. Use an API key for agent workloads and add backoff on 429s. On self-hosted instances, tune API_RATE_LIMIT for your expected agent traffic.