Solana AI Kit plugin
Agents, commands and MCP for Solana work in Claude Code, plus a firewall that fences secrets and the main network.
- 111 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The installer is downloaded over the network, run through `bash` and rewrites agent settings: `.claude/`, hooks, `.mcp.json`.
- Kit commands deploy programs and work with Solana wallets; protection rests on the firewall tier and the `safe-ai-skill` plugin, and on lower tiers network writes are only confirmed.
Install
- Works in
Manual install
curl -fsSL https://aikit.superteam.codes | bashThe first command from the README. Run it from the project root; afterwards run `claude` and `/setup-mcp`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The installer is downloaded over the network, run through `bash` and rewrites agent settings: `.claude/`, hooks, `.mcp.json`.
- Kit commands deploy programs and work with Solana wallets; protection rests on the firewall tier and the `safe-ai-skill` plugin, and on lower tiers network writes are only confirmed.
This is third-party code. Review the repository files before installing.
What it does
Solana AI Kit puts a Claude Code setup for Solana development into the project: 15 agents (architect, Anchor, Pinocchio, DeFi, tokens, frontend, mobile, QA and others), 54 commands, skill packs from Solana Foundation, Jupiter, Metaplex, Helius and more, and four MCP servers for chain data through Helius, Solana docs, library docs and context compression. Almost nothing loads into context until a task needs it. A separate layer protects the project. A firewall with the Off, Relaxed, Medium and High tiers limits secret reads, destructive commands and on-chain writes, and the `safe-ai-skill` plugin checks money operations and keeps transfers within limits. For Codex, opencode and other agents that read `AGENTS.md`, installation uses the `--agents` flag.
Who it is for. Developers of Solana programs and apps who want an agent that knows Anchor, Pinocchio and client libraries, plus protection against an accidental deploy to the main network.
Good fit when
- An Anchor project is starting and one setup should carry an architect and a QA agent.
- You want to measure a program's compute-unit use with `/profile-cu` before and after a change.
- A user transaction has to be traced and a token mint inspected.
- An agent works in a repository with wallets, and secrets and deploys must be fenced by the firewall.
- You want to run the whole path from idea to PR with the `intake`, `plan-work` and `ship` commands.
Not a fit when
- The project is not about Solana: the kit is built entirely for that network.
- Running a downloaded installer is forbidden and there is no time to read a clone file by file.
- Work happens on Windows without WSL: the safety plugin blocks every tool call there, and the installer skips it.
- You need a short rules file: the install adds dozens of agents and commands.
Example request
Run `/audit-solana` for the program in `programs/vault` and show the problems found.Limitations
The installer is downloaded over the network and run through `bash`, then creates `.claude/`, `CLAUDE.md`, `.mcp.json`, `.env` and `.safe-ai-skill/`; you can download and read it before running. Chain data needs a Helius key, which the `/setup-mcp` command enters. Firewall hooks work only in Claude Code and are absent in installs for Codex and opencode. One external pack, `solana-new`, after its own setup sends the skill name, version and system; it is turned off with `{"telemetryTier":"off"}` in `~/.superstack/config.json`.
How to disable. Turn the safety plugin off with `"safe-ai-skill@stbr": false` under `enabledPlugins` in `.claude/settings.local.json`. Remove the rest by hand: `.claude/`, `.mcp.json`, `.safe-ai-skill/`, the added `CLAUDE.md` blocks and Helius keys from `.env`.
FAQ
How do I install Solana AI Kit?
From the project root run `curl -fsSL https://aikit.superteam.codes | bash`, then start `claude`. To read the script first, download it to a file and run the copy you read.
How do I install the kit for Codex or opencode?
Add a flag: `curl -fsSL https://aikit.superteam.codes | bash -s -- --agents`. Agents read `AGENTS.md` and `.agents/skills/`, but firewall hooks do not work in this mode.
What firewall tiers exist?
Four: Off, Relaxed by default, Medium and High. The `/firewall` command picks the tier; on High writes to the main network are denied outright, and on Relaxed they need confirmation.
How do I connect chain data?
Run `/setup-mcp` in Claude Code. The command walks you through the Helius key and offers optional MCP servers, showing what each one exposes.
How do I update the kit?
Run `/update` in Claude Code: it takes the latest release. The `bash .claude/bin/update.sh` script refreshes skill packs from `ext/`.
Discussion
No questions or reviews yet.
Related
PluginsQuicknode plugin and skill for Solana programs in Anchor 1, 2 and Quasar: accuracy, LiteSVM tests and compact Rust.
A Solana Foundation skill for Solana development: Anchor and Pinocchio, @solana/kit clients, wallets, Surfpool and LiteSVM testing, security
A set of 116 skills, 44 agents and safety hooks for Claude Code, Cursor, Codex, Gemini CLI and nine more editors
Over twenty skills for coding agents: MCP and hooks, plan mode, a bug-fix protocol, prompting and a bash safety hook
Claude Code kit with eight phase agents, 21 skills and hooks that keep the agent inside set limits.
A set of Claude Code utilities: background skills, expert agents, slash commands and prompt templates for development
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Solana AI Kit (this card) | no | 111 | |
| Solana Development Skill | yes | 590 | a standalone Solana Foundation skill for Anchor, Pinocchio and tests; Solana AI Kit adds agents, 54 commands, MCP servers and a firewall |
| Claude Forge | no | 879 | a general Claude Code build of agents, commands, hooks and MCP for many projects; Solana AI Kit is tailored to one network and limits deploys and secrets |
README in short
The README explains what the kit contains, gives three ways to install, describes the four firewall tiers and the memory and code-intelligence plugins. Details sit in `docs/`: agents and commands, skill packs, configuration and a comparison of agents.