Gives an AI coding agent the ability to build on GetBlock, not just describe it. The agent scaffolds runnable projects, provisions endpoints, wires in risk screening, estimates cost, and diagnoses failures — by running scripts that ship with the skill.
npx skills add https://github.com/GetBlock-io/getblock-blockchain-skills --skill getblock-skillThe skill expects GetBlock's MCP server. With it connected, the agent reads your plan, tokens, and the live chain catalogue, and creates endpoints for you instead of sending you to the dashboard.
claude mcp add --transport http getblock https://mcp.getblock.io/ \
--header "Authorization: Bearer <GETBLOCK_API_KEY>"Use the account API key from the GetBlock dashboard, not a node access token. Other clients take the same URL and header in their MCP config. Without the server, the skill still scaffolds, estimates, and diagnoses, but it cannot see your account or provision endpoints.
Works with any agent supporting the skills framework — Claude Code, Cursor, Codex, OpenCode, Windsurf. The skill loads automatically when the agent starts blockchain work.
| Workflow | What actually happens |
|---|---|
| Scaffold a project | Runs setup_project.py — emits a runnable viem / ethers / Hardhat / Foundry / web3.py / Solana project wired to GETBLOCK_RPC_URL |
| Provision an endpoint | Reuses a matching token from list_tokens, or calls get_protocol to confirm what the chain supports, confirms with you, then create_token and writes the URL into .env |
| Add risk screening | Generates a working pre-transaction gate against Address Audit, with the string/hard-override/new-address traps already handled |
| Screen an address now | Picks the right endpoint of five by what is being screened, and calls it |
| Estimate cost | Runs cu_estimate.py against the published multiplier tables |
| Diagnose a failure | Runs verify_setup.mjs, checks the token and its real endpoint with list_tokens, maps the result onto GetBlock's error taxonomy |
Covers Shared / Limitless / Dedicated nodes, Compute Units, Crypto AML, Wallet Risk, Wallet Audit, Rug Pull Checker, Solana Market Data, Shreds Streaming, Yellowstone gRPC, BSC Advanced Tooling, TRON Energy, the GetBlock MCP server, and both extension catalogues — the six blockchain-specific add-ons and the nine extra services for dedicated nodes.
"Using GetBlock skills, set up a viem project on Base."
-> scaffolds the project, provisions an endpoint if needed, verifies the
connection, and leaves you with something that runs.
"Using GetBlock skills, block sanctioned wallets before my checkout signs."
-> generates the screening gate and wires it into the transaction path.
"Using GetBlock skills, what will it cost to poll eth_blockNumber every
second on Polygon?"
-> runs the estimator and compares it against every plan tier.
"Using GetBlock skills, my endpoint returns 401 Unknown token."
-> runs the diagnostic and checks the region before touching the token.
skills/getblock-skill/
├── SKILL.md nine workflows, each ending in something that runs
├── PRODUCT_CATALOG.md maintenance source of truth, coverage status, known docs issues
├── scripts/
│ ├── setup_project.py scaffolds a runnable project for 6 frameworks, 11 chains
│ ├── cu_estimate.py Compute Unit cost estimator
│ └── verify_setup.mjs endpoint + product-key diagnostic
└── references/ depth, loaded on demand
├── endpoints-auth-reference.md URL shape, regions, token scope, archive mode
├── compute-units-reference.md CU formula, multiplier tables, cost reduction
├── plans-reference.md Shared vs Limitless vs Dedicated, sizing, billing
├── errors-reference.md HTTP and JSON-RPC taxonomy, diagnosis order
├── risk-compliance-reference.md all five risk endpoints, request/response shapes
├── risk-scoring-reference.md score polarity, bands, hard override
├── risk-product-selection.md which of the five endpoints to call
├── risk-integration-patterns.md where in a flow to screen
├── solana-market-data-reference.md
├── streaming-reference.md Shreds, Yellowstone, BSC tooling
├── tron-energy-reference.md
├── addons.md 6 add-ons, 9 extra services, what each plan can have
├── admin-api-reference.md GetBlock MCP server: connecting, tools, provisioning rules
└── gotchas.md
Auth is not one model. Node access puts the token in the URL path
(https://shared.<region>.getblock.io/<ACCESS_TOKEN>/) and cannot use a header. The
on-demand products use a bearer header on services.getblock.io, and use thesame API KEY obtain from dashboard.
A Compute Unit is not a request. Total CU = chain × method × archive. The cheapest
Ethereum call costs 20 CU, so 50,000 CU/day is 2,500 calls — and 166 on Moonbeam, where
the chain multiplier is 300.
PRODUCT_CATALOG.md is the source of truth for what is covered and how well. Update the
reference file first, the SKILL.md section only if a headline fact or setup snippet
changed, then bump last_verified.
Plan prices, limits, and multiplier tables are a dated snapshot — verify against
https://getblock.io/pricing/ before quoting them to a customer.
