Superteam Brazil Relaunches Solana-Claude as Solana AI Kit, Available Now on the Claude Code Plugin Marketplace

Superteam Superteam Brazil shipped a significant rebranding and expansion of its open-source Solana developer toolkit on June 17. The project, previously known as "solana-claude," is now Solana AI Kit and is available directly from the Claude Code plugin marketplace, meaning any developer running Claude Code can install it with a single command.

The kit aggregates 15 specialized agents, 29 workflow commands, 18 external skill submodules, and 7 MCP server integrations into a single Claude Code configuration targeting the full Solana SOL$76.400.0% Solana development lifecycle, from program architecture through deployment, security auditing, and documentation.

What Changed: Plugin Marketplace Distribution

The most consequential shift is distribution. Solana AI Kit now installs as a Claude Code plugin:

/plugin install solana-ai-kit@stbr

Or via the one-liner:

curl -fsSL https://aikit.superteam.codes | bash

That install endpoint (aikit.superteam.codes) is live and resolves to the GitHub repository at github.com/solanabr/solana-ai-kit. Making the toolkit available on the Claude Code marketplace gives it visibility to any developer who browses plugins, not just those who find the Superteam Brazil GitHub org.

15 Agents Across the Full Stack

The kit's 15 agents are domain-scoped rather than general-purpose. Each agent carries a focused system prompt and a narrowly scoped skill set:

• solana-architect: system design, program derived account (PDA) schemes, and multi-program architecture
• anchor-engineer: Anchor 1.0 development, IDL generation, and account constraints
• pinocchio-engineer: compute unit (CU) optimization via zero-copy and minimal binary approaches, with the README claiming 80–95% CU savings
• defi-engineer: DeFi protocol integrations covering Jupiter, Drift, Kamino, Raydium, Orca, and Meteora
• token-engineer: Token-2022 extensions and launch mechanics
• mobile-engineer and solana-frontend-engineer: React Native/Expo and Next.js respectively
• rust-backend-engineer: Axum APIs, indexers, and WebSocket services
• devops-engineer: CI/CD, RPC infrastructure, and Cloudflare Workers
• solana-qa-engineer: testing (Mollusk, LiteSVM, Trident), CU profiling, and code quality
• game-architect and unity-engineer: Unity/PlaySolana/PSG1 game development
• tech-docs-writer, solana-guide, and solana-researcher: documentation, tutorials, and research

The multi-agent design lets Claude Code orchestrate sequential agent teams for complex workflows (architect, then anchor-engineer, then qa-engineer, then tech-docs-writer) without the developer needing to manually switch contexts.

29 Commands, Including /audit-infra and /doctor

The command set is the most immediately useful layer for most developers. Highlights from the 29 commands:

/doctor runs a health check on the development environment, validating RPC endpoints, API keys, MCP server connectivity, and toolchain state. It is the first command the README recommends running after installation.

/audit-infra performs an infrastructure-first security audit covering secrets management, supply chain risks, CI/CD pipeline security, LLM and skill security, OWASP vulnerability categories, and STRIDE threat modeling. The scope is meaningfully different from /audit-solana, which covers on-chain program logic. Infra auditing catches risks that never touch Solana accounts at all: leaked keys in CI logs, compromised dependencies, insecure environment configurations.

/profile-cu and /benchmark support CU profiling per instruction with before-and-after comparison, addressing one of the more time-consuming parts of Solana program optimization.

/dream consolidates and deduplicates agent memory across sessions, preventing context drift in long-running development workflows.

Other commands cover building (/build-program, /build-app, /build-unity, /scaffold), testing (/test-rust, /test-ts, /test-dotnet), deployment (/deploy), migration (/migrate-web3 for @solana/web3.js → @solana/kit), and workflow tooling (/quick-commit, /setup-ci-cd, /setup-mcp).

18 External Skill Submodules From Ecosystem Partners

The 18 external skill submodules are the most architecturally distinctive element. Rather than building every capability in-house, the kit pulls from official repositories maintained by the ecosystem organizations that know their systems best. The full kit counts at a glance:

The confirmed submodule list, from .gitmodules in the repository:

The Anthropic "defending-code" submodule is notable: it brings Anthropic's own vulnerability-discovery harness into the Solana development environment, extending the security toolkit beyond Solana-specific risks to general software weaknesses.

Token-Efficient Architecture

The kit's design philosophy centers on reducing context overhead. The main CLAUDE.md is delivered as a user message rather than a system prompt, preserving the system prompt budget for other tools. Agent-specific skills load progressively based on file patterns detected in the working directory, rather than pre-loading everything at startup. This matters in practice: a developer working on a React frontend will not have the Anchor or Rust-backend agent contexts loaded unless they open .rs or Anchor-specific files.

The 7 pre-configured MCP servers follow a similar selectivity principle. Per the kit's README, Helius provides 60+ tools covering RPC, DAS API, webhooks, priority fees, token metadata, and NFT data. Surfpool handles agent-driven local validator and mainnet-fork control. The remaining five servers cover live library documentation (Context7), browser automation for dApp testing (Playwright), persistent semantic search across sessions (memsearch), context-window compression for RPC responses and build logs (context-mode), and the Solana Foundation's official docs and API reference (solana-dev MCP).

Marketplace Discovery and the Developer Funnel

Solana's developer toolchain has matured substantially in 2026. Earlier this month, Triton One open-sourced SuperBank, a 38x faster historical ledger layer on ClickHouse, and Solana Mobile launched ratings, reviews, and AI summaries for its dApp Store Publishing Portal. Solana AI Kit extends that maturation into the AI-native tooling layer.

The kit being on the Claude Code plugin marketplace specifically broadens the developer funnel. Developers who arrive at Claude Code without prior Solana context can discover a production-ready Solana configuration rather than assembling one from scratch. The /doctor health check and guided install flow reduce time-to-first-compile for new builders.

For existing Solana developers, the practical case is /audit-infra alongside the Trail of Bits and Ghost Security submodules. CI/CD hygiene, secret scanning, and supply-chain checks tend to get skipped when teams are shipping fast. Having them as a one-command operation inside the same AI assistant used for development lowers the activation energy considerably.

Superteam Brazil's decision to route the install domain through aikit.superteam.codes as a permanent redirect to the GitHub installer also keeps the install instruction stable: any documentation pointing to that URL will continue to work regardless of repository reorganization.