A raw event log is hex-encoded topics and data. The SQD Portal MCP server hands an agent pre-decoded events, transactions, traces, and fills as typed JSON across 225+ networks through one endpoint, so the model reads onchain state instead of decoding it.
Events arrive with decoded parameter names and typed values. No ABI lookup, no hex-to-int conversion. Your agent gets {"from": "0x...", "to": "0x...", "value": "27269390699"}, not raw log topics.
Feed the agent new blocks, events, and state changes as a continuous NDJSON stream over one HTTP connection. Its context stays current without polling or cron jobs.
The same query shape works across Ethereum, Arbitrum, Base, Polygon, and Solana. An agent reuses one integration and one tool instead of five RPC clients with five different schemas.
Add the MCP server to Claude Code, Cursor, or VS Code, or call the same Stream API over a plain HTTP POST from any agent framework. No SDK lock-in, no client library required.
A raw Ethereum event log is a blob of hex-encoded topics and data fields. An LLM needs to know the ABI, decode the parameters, and parse binary data into typed values, all before it can start reasoning about what actually happened.
That is context-window budget burned on data wrangling for every single event. Multiply by hundreds of events per block across multiple chains, and a large share of the agent's context goes to decoding instead of reasoning.
SQD delivers pre-decoded, typed JSON with human-readable field names. Your agent gets structured data it can reason about immediately.
A real row from the Portal MCP server. value is in base units: with 6 decimals that is 27,269.39 USDC. Verify it at 0x4933…86b1. No ABI lookup, no hex conversion.
Your agent speaks to the MCP server. The MCP server calls Portal. Portal serves the data from a network of 2,000+ worker nodes. Your agent gets typed JSON it can reason about.
"Get all USDC transfers over $1M on Base in the last hour"
portal_stream → base-mainnet → ERC-20 Transfer filter
POST /datasets/base-mainnet/stream → 2,000+ worker nodes
decoded Transfer rows · from, to, value · typed JSON
The endpoint is https://portal.sqd.dev/mcp. In Claude Code, one command adds it:
Cursor and VS Code take the same URL in their MCP config. One thing to tell the agent up front: heavy queries come back as a bounded preview page with a cursor, so scope by block range and page through _pagination.next_cursor. The full setup, with three worked queries, is in the guide.
Watch how the MCP server lets your agent query onchain data through natural language, no SDK setup required.
LLM-optimized docs, OpenAPI schemas, and skill definitions, everything your agent needs to understand SQD without reading HTML.
Live queries against real blockchain data. Pick a network, choose a query, hit run.
The public endpoint is free for development. See plans for dedicated portals and higher limits.
Each of these runs on the same SQD data layer.
A DEX swap settles its token movement in nested internal calls; an event-log-only feed sees the receipt, not the execution.
SQD sits at the extract-and-decode end of the analytics pipeline.
Query any address and get its history: token transfers, native and internal value, NFT movements, and contract calls, across 225+ networks through one query shape.
Onchain, every stablecoin movement is a Transfer event: a mint is a transfer from the zero address, a burn is a transfer to it, a payment is one between two wallets.
A tokenized treasury fund is a regulated security that lives on a public chain.
Most indexers give you event logs.
Wire the SQD Portal MCP server into Claude Code, Cursor, or VS Code, then run three real onchain queries and read the rows back.
An agent answers whole-market Hyperliquid perp questions through the same analytics, OHLC, and query tools it uses for every other chain.
What an agent actually needs from onchain data, the Model Context Protocol (MCP) pattern, and how on-demand and pre-indexed pipelines differ in agent workflows.
sqd CLI. Agents can scaffold a new Squid SDK project, generate entities and batch-processor logic, and run sqd deploy to push to SQD Cloud, all coordinated through natural language in their host environment (Claude Code, Cursor, and similar). The MCP server itself is read-only against Portal; deployment to Cloud goes through the standard sqd CLI tool.Private Portal. Dedicated. Validated. Managed. Tell us what you're building, we'll show you what it looks like on SQD.