For Docs
- Docs AI Search - Search documentation with AI in the search bar.
- Markdown Export - Append
.mdto any page URL for raw markdown.
- View
/llms.txtfor an index of the docs - It lists key pages with descriptions so agents can navigate to answers quickly.
For Development
Agent Skills
View or install/skill.md, a structured capability file that tells agents what they can do
with Light Protocol and ZK Compression. If you’re building with agents, start here.
| Use case | Skill |
|---|---|
| Build DeFi programs (AMMs, vaults, lending) with Anchor or Pinocchio | defi-program |
| Integrate rent-free markets into routers and aggregators | defi-router |
| Stream account state via Laserstream gRPC | data-streaming |
| Wallets and payment flows with light-token. Includes privy, wallet adapter, mobile wallet adapter signing. Optional nullifier to prevent your onchain instruction from being executed more than once. | payments-and-wallets |
| Airdrops, DePIN, token distribution | airdrop |
| Anti-double-spend nullifiers for Privacy-preserving ZK programs | zk-nullifier |
| Testing programs and clients on localnet, devnet, mainnet | testing |
| Help with Debugging and Questions via DeepWiki MCP | ask-mcp |
View all skills here: https://github.com/Lightprotocol/skills.All skills are included and are auto-discovered based on context. Ask about light-token, defi, payments, or program migration and the agent uses the relevant skill automatically.
- Claude Code
- Cursor
- Any Agent
Add the marketplace and install:
MCP
The Model Context Protocol (MCP) is an open standard to connect AI apps to data sources and tools. The DeepWiki MCP server provides access to the Light Protocol repository with its search capabilities (AskDevin).Installation
- Claude Code
- Codex
- Most Clients (Windsurf, Cursor, ...)
DeepWiki MCP-Tools
- read_wiki_structure - Get a list of documentation topics for a GitHub repository
- read_wiki_contents - View documentation about a GitHub repository
- ask_question - Ask any question about the GitHub repository and get a context-grounded response