AI coding tools for Solana compared

Short answer

General coding agents such as Claude Code, Codex CLI and Cursor work in any language, and the Solana Foundation’s Developer MCP and solana-dev skill add Solana documentation to them. Oh My Sol is a Solana harness that runs the model you already use with Solana references, error decoding, toolchain checks, a local Surfpool chain and a check that closes a task only when your requirements pass. It beats both Codex CLI and Claude Code on Solana Bench and fresh Solana tasks.

Who it’s for
Solana developers choosing an AI coding setup. Barestate makes Oh My Sol; other tools are described from their own documentation.
Last checked
, against Oh My Sol 0.1.0

Which AI coding tools work for Solana?

The tools fall into three groups. General coding agents do the work in any language. Add-ons give those agents Solana documentation. Solana-specific products build the whole workflow around Solana. Barestate makes Oh My Sol; everything said about the other tools comes from their own sites and docs.

ToolTypeRunsLocal chain or checkModel
Oh My Sol (OMS)Solana harnessYour terminal (macOS, Linux, WSL 2)Built-in Surfpool chain. A task closes only when its requirements pass.The model you already use: a ChatGPT or Codex subscription, a Claude subscription through Claude Code, or an API key.
Claude CodeGeneral coding agentTerminal, IDE extensions, desktop app, webNot built in. It runs whatever local validator or tests your project already uses.Claude, with a Claude subscription or an Anthropic Console account.
Codex CLIGeneral coding agentTerminalNot built in. It runs whatever your project already uses.OpenAI models, signed in with ChatGPT or another supported sign-in method.
CursorAI code editorDesktop editorNot built in.The models available in Cursor.
Solana Developer MCP and the solana-dev skillAdd-on for any agentRemote MCP server; skill files in your agentNone. The skill documents Surfpool and LiteSVM testing for your own setup.Whatever agent you add them to. The MCP server needs no API key.
CódigoCloud IDE for SolanaBrowserEach workspace has a solana-test-validator and an auto-generated wallet.Its own custom-trained model.

What does each tool do?

Oh My Sol (OMS)

Barestate · Solana harness

A terminal app that runs the model you already use on Solana work. Each task gets its own Git worktree. The model can only propose that it is done, and a separate check runs your requirements on a fresh offline Surfpool chain before the task closes.

Best for
Writing, testing and debugging Solana programs, with the work checked on a local chain before it counts as done.
Solana knowledge
An offline reference for 50 programs and 1,559 instructions, a pinned Solana documentation set, program error decoding, toolchain checks, and the official Solana MCP server when web access is on.
Local chain or check
Built-in Surfpool chain. A task closes only when its requirements pass.
Model
The model you already use: a ChatGPT or Codex subscription, a Claude subscription through Claude Code, or an API key.

Claude Code

Anthropic · General coding agent

Anthropic’s agentic coding tool. It reads your codebase, edits files, runs commands and works in any language. You can extend it with MCP servers, skills and hooks.

Best for
General coding in any language, and Solana work once you add Solana tools to it.
Solana knowledge
The model’s own knowledge, plus MCP servers and skills you add, such as the Solana Developer MCP.
Local chain or check
Not built in. It runs whatever local validator or tests your project already uses.
Model
Claude, with a Claude subscription or an Anthropic Console account.

Codex CLI

OpenAI · General coding agent

OpenAI’s terminal coding agent. It inspects a local repository, edits files and runs the tools on your machine, with permission settings for when it can act without asking, optional live web search and MCP support.

Best for
Terminal-first coding in any language, scripting and CI, and Solana work once you add Solana tools to it.
Solana knowledge
The model’s own knowledge, optional web search, and MCP servers you add with codex mcp add.
Local chain or check
Not built in. It runs whatever your project already uses.
Model
OpenAI models, signed in with ChatGPT or another supported sign-in method.

Cursor

Cursor · AI code editor

An AI code editor whose Agent can search your codebase, edit files and run terminal commands. It connects to MCP servers configured in a project or global mcp.json file.

Best for
Developers who want AI help inside an editor rather than a terminal.
Solana knowledge
The model’s own knowledge, plus MCP servers you add, such as the Solana Developer MCP.
Local chain or check
Not built in.
Model
The models available in Cursor.

Solana Developer MCP and the solana-dev skill

Solana Foundation · Add-on for any agent

The MCP server gives an agent Solana documentation retrieval, semantic search, an ask-for-help tool and program_autofixer, which checks Anchor and Pinocchio program Rust for security antipatterns. The skill is an MIT-licensed set of instructions and references for Solana development.

Best for
Keeping the agent you already use and giving it Solana documentation and patterns.
Solana knowledge
Solana documentation through the MCP server, and the skill’s guidance on Anchor, Kit, Surfpool, LiteSVM, Mollusk and security.
Local chain or check
None. The skill documents Surfpool and LiteSVM testing for your own setup.
Model
Whatever agent you add them to. The MCP server needs no API key.

Código

Código · Cloud IDE for Solana

A browser IDE for Solana with its own custom-trained model. It generates Anchor or native programs, tests and frontends from a prompt, with templates and one-click deploy.

Best for
Getting a Solana program running in a browser without a local toolchain.
Solana knowledge
Its own model trained for Solana, plus templates for Anchor and native programs.
Local chain or check
Each workspace has a solana-test-validator and an auto-generated wallet.
Model
Its own custom-trained model.

How does Oh My Sol score against Claude Code and Codex CLI?

OMS beats both. Solana Bench scores how many distinct Solana instructions an agent executes successfully on a mainnet fork. The fresh tasks are eight Solana tasks, each graded by a hidden check.

Solana Bench, mean scoreOh My SolOwn harnessOMS lead
GPT-6.1 Sol, basic (vs Codex CLI)377273+38%
GPT-6.1 Sol, swap (vs Codex CLI)366283+30%
Claude Opus 5.5, basic (vs Claude Code)303252+20%
Claude Opus 5.5, swap (vs Claude Code)258230+12%
Claude Sonnet 5.5, basic (vs Claude Code)208100+107%
Claude Sonnet 5.5, swap (vs Claude Code)17693+90%

Each model ran in Oh My Sol and in its own harness on the same Solana Bench environment.

Fresh tasks, same model in bothOh My SolOther harness
GPT-6.1 Sol, against Codex CLI96%76%
Claude Opus 5.5, against Claude Code92%79%
Claude Sonnet 5.5, against Claude Code71%46%

OMS also used 42 to 62% fewer input tokens for every task it solved. The benchmarks overview has every chart.

How do I choose?

  • You are writing, testing or debugging a Solana program. Use OMS with the model you already pay for. It keeps Solana references, error decoding and a local chain next to the model, and the task closes only when your requirements pass.
  • You already use Claude Code, Codex CLI or Cursor for other work and want Solana docs in it. Add the Solana Developer MCP and the solana-dev skill. Both are free.
  • You want a browser IDE with no local setup. Código is built for that.

Can I combine these tools?

Yes. The MCP server and the skill work alongside any agent that supports them, and OMS already includes some of this: when web access is on it calls the official Solana MCP server, and its built-in reference pack is built from the MIT-licensed solana-dev skill and two Helius skills. OMS can also run your Claude subscription through Claude Code or your ChatGPT subscription through Codex, so you keep the model you already use. Many people keep an editor such as Cursor open and run OMS in a terminal next to it.

Frequently asked questions

What is the best AI coding tool for Solana development?

Oh My Sol (OMS) is built for it. It runs the model you already use, adds Solana references, error decoding, toolchain checks and a local Surfpool chain, and closes a task only when your requirements pass. It beats both Codex CLI and Claude Code on Solana Bench and on fresh Solana tasks. If you prefer to stay in a general agent or editor, add the Solana Developer MCP and the solana-dev skill to it.

How does Oh My Sol compare with Claude Code and Codex CLI?

On Solana Bench, OMS averaged 377 in the basic environment against 273 for Codex CLI on GPT-6.1 Sol, and 303 against 252 for Claude Code on Claude Opus 5.5. On eight fresh Solana tasks it passed 96% of runs against 76% for Codex CLI with GPT-6.1 Sol, and 92% against 79% with Claude Opus 5.5 and 71% against 46% with Claude Sonnet 5.5 for Claude Code, using 42 to 62% fewer input tokens per solved task.

Can I use the Solana Developer MCP with Claude Code, Codex or Cursor?

Yes. Its setup page gives one-line instructions for Codex, Claude Code, Cursor, Windsurf and VS Code, and it needs no API key. OMS also calls it when web access is on.

Which of these tools runs a local Solana chain?

OMS runs a local Surfpool chain and checks finished work on a fresh copy of it before a task closes. Código workspaces include a solana-test-validator. The general agents use whatever validator or tests your project already has, and the MCP server does not run a chain.

Who wrote this comparison?

Barestate, which makes Oh My Sol. Facts about the other tools come from their own sites and documentation. If something is out of date, tell us through the contact page and we will fix it.

Try it on your own program

Oh My Sol 0.1.0 runs on macOS and Linux, and on Windows through WSL 2. You need Node 22.19.0 or newer and Git. You bring your own model access: an API key, a ChatGPT subscription or a Claude subscription.

npx @barestate/oms doctor --install
npx @barestate/oms