Restrict RPC Access (the read-side of access privacy)

Overview

Allowlist precompiles gate writes (txs, deploys). To make an L1 truly private you must also gate reads — i.e. restrict the RPC endpoints so only authorized participants can query state. This is the "walled garden" half of access privacy ([[private-permissioned-l1s]]).

Layers of read restriction

1. Don't expose public RPC. Run validators/nodes on a private network (VPC/VPN); do not publish a public RPC URL or list it on chain explorers.
2. Validator-only / private nodes. Keep RPC bound to internal interfaces; provide RPC only to authorized API nodes behind authentication.
3. Network/auth layer. Put RPC behind a reverse proxy/gateway requiring API keys, mTLS, or VPN. The chain protocol doesn't authenticate RPC callers — your infra does.
4. Node config. Configure the avalanchego/Subnet-EVM node's HTTP/API settings (enabled APIs, host binding, allowed origins) to minimize the surface. Per-L1 chain configs live under the node's chain-config directory.

Combine for full privacy

Goal	Mechanism
Who can read	Restricted RPC (this skill) + private networking
Who can transact	Transaction AllowList ([[l1-transaction-allowlist]])
Who can deploy	Deployer AllowList ([[l1-deployer-allowlist]])
Who can validate	ValidatorManager / PoA ([[poa-validator-manager]])
Data confidentiality	eERC on top ([[combining-eerc-with-private-l1]])

Operational notes

• Treat RPC like any sensitive API: authenticate, rate-limit, log, and monitor.
• For consortiums, give each member their own credentialed RPC gateway rather than a shared public endpoint.
• Even with restricted RPC, on-chain data is plaintext to anyone who does have access — add eERC if members shouldn't see each other's amounts.

Common Pitfalls

• Allowlists without RPC restriction. A common mistake — writes are gated but anyone can still read all state. Not private.
• Public explorer indexing. If you ever expose RPC publicly, indexers may cache your data permanently.
• Assuming the protocol authenticates RPC. It doesn't; access control is an infrastructure responsibility.

AI Agent Prompt

"Act as an infrastructure engineer. Design a private Avalanche L1 where only consortium members can reach the RPC: private networking, authenticated RPC gateways, hardened node API config, plus tx/deployer allowlists and a PoA validator set."

References

See references/README.md (academy permissioned-l1s 'Private L1s', builders-hub node chain-configs, access-restriction course).