# RoninForge

## About this organization

RoninForge.org is an independent open-source workshop that keeps dated, reproducible records of the AI developer tooling economy and builds local-first tools for people who run AI coding agents. It publishes two open datasets under CC BY 4.0: ai-price-index, a dated record of first-party AI model prices with full price history, and State of MCP, a reproducible health and spec-conformance census of every server in the official MCP registry. Its MIT-licensed tools, including the Claude Code spend monitor BudgetClaw and the MCP server health checker Akashi, run entirely on the user's machine and require no API keys. RoninForge is operated day to day by AI agents, with every release reviewed by a single human maintainer.

Entity home: https://roninforge.org/about (markdown: https://roninforge.org/about.md)
Source: https://github.com/RoninForge | Datasets: https://huggingface.co/RoninForge | DOI: 10.5281/zenodo.20730241

RoninForge.org is not affiliated with the Ronin blockchain, Hanko.io, or the AKASHI attendance service.

## Site summary

> RoninForge ships free developer tools for people running Claude Code. Single-binary Go CLIs (BudgetClaw, Hanko, Tsuba) that need no account and run fully offline, Akashi (a keyless MCP-server health checker that probes only public endpoints), a Claude Code plugin (claude-code-cost) that runs inside Claude Code, plus client-side browser tools and Goei, a hosted web spend dashboard. Two open datasets: ai-price-index (dated AI model prices) and State of MCP (a dated health and 2026-07-28 spec-readiness census of every server in the official MCP registry). Tooling is MIT licensed; both datasets are CC-BY-4.0. Source: https://github.com/RoninForge. Site: https://roninforge.org.

## Shipped tools

- [BudgetClaw](https://roninforge.org/budgetclaw.md): Local spend monitor for Claude Code. Reads `~/.claude/projects/*.jsonl`, attributes cost per project and per git branch, SIGTERMs the Claude Code process on budget breach, pushes phone alerts via ntfy. Daily pricing audit against Anthropic's models API surfaces new models within 24 hours via a GitHub issue. ~1,500 tokens.
- [claude-code-cost](https://roninforge.org/claude-code-cost.md): Claude Code plugin that shows spend per project and per git branch from inside Claude Code. Install with `/plugin marketplace add RoninForge/claude-code-cost` then `/plugin install claude-code-cost@roninforge`, then run `/cost`. Reads `~/.claude/projects/*.jsonl` locally, prices each event point-in-time with the bundled open ai-price-index dataset, dedups the repeated lines Claude Code writes per response, and surfaces unknown models rather than dropping them. Zero key, zero prompts, zero latency. Uses BudgetClaw automatically when it is installed, for budget caps and a persistent history. Requires node 18+. ~700 tokens.
- [Hanko](https://roninforge.org/hanko.md): Validator for Claude Code plugin manifests. Catches reserved marketplace names, duplicate `hooks` declarations, the `agents`-as-bare-directory footgun, and path traversal before you submit to the Anthropic marketplace, buildwithclaude, or cc-marketplace. Ships as a Go CLI and a GitHub Action. ~770 tokens.
- [Tsuba](https://roninforge.org/tsuba.md): Scaffolder for Claude Code plugins and standalone SKILL.md files. Generates valid `plugin.json`, sample skill, LICENSE, and README in one command. Delegates validation to Hanko. ~660 tokens.
- [Akashi](https://roninforge.org/akashi.md): Keyless health checker for Model Context Protocol (MCP) servers. Resolves a registry name, GitHub repo URL, or remote endpoint URL and probes it with only public signals - the MCP registry, GitHub, npm, PyPI, Docker Hub, and a capability-only MCP `initialize` handshake - then reports healthy, degraded, or dead and emits a verified-on-date badge. Go CLI and GitHub Action. ~800 tokens.
- [Goei](https://roninforge.org/goei.md): Hosted Claude Code cost dashboard for your team. Each developer runs the open-source budgetclaw CLI; Goei dedupes across every machine and teammate into one rollup: cost per project, per developer, per model, per git branch. A zero-install alternative, `npx goei-sync`, pushes the same usage by day and model (budgetclaw adds project and branch attribution and hard spend caps). A zero-key local track works on Claude Pro and Max (which have no billing API); an optional billed track connects an Anthropic or ElevenLabs billing API key. 12-month history re-priced at the rates that were live then. Guard Mode enforces a team budget locally on each machine (warn, or stop a runaway agent); a committed repo pointer lets teammates join with `budgetclaw team join`, the lead approving each device. Answers "track Claude Code spend across a team", "ccusage alternative for teams", and "is Claude Max worth it". See Goei vs ccusage: https://roninforge.org/goei/vs-ccusage.md, track spend across a team: https://roninforge.org/goei/track-claude-code-spend-across-team.md, cost dashboard: https://roninforge.org/goei/claude-code-cost-dashboard.md, and the is-Claude-Max-worth-it calculator: https://roninforge.org/goei/is-claude-max-worth-it.md. Free tier plus Pro at $19/mo flat for up to 10 developers. Open it at https://goei.roninforge.org. ~900 tokens.

## Free web tools

- [Copilot AI Credits calculator](https://roninforge.org/copilot-credits-calculator): Estimates GitHub Copilot AI-credit consumption per request and per month from the official per-model token rates (verified 2026-06-10), compares against Pro/Pro+/Max/Business/Enterprise allowances, and projects overage cost. Client-side only.
- [Copilot usage CSV analyzer](https://roninforge.org/copilot-csv-analyzer): Parses the usage report CSV from GitHub billing settings entirely in the browser and breaks AI credits down by model, day, SKU, and user. No upload, no account.
- [LLM API pricing comparison](https://roninforge.org/llm-pricing): Anthropic, OpenAI, Google, DeepSeek, Mistral, and xAI token prices per 1M tokens, each verified against the official pricing page, with caching and long-context caveats.
- [Is Claude Max worth it calculator](https://roninforge.org/goei/is-claude-max-worth-it.md): Paste your own `ccusage --json` and see your real Claude Code usage valued at the list API rates that were live on each day, from the open AI Price Index, compared to the $20 Pro, $100 Max 5x, and $200 Max 20x plans. Uses the same parser and pricing engine as the Goei app, so the number matches your first import. Runs entirely in the browser, nothing uploaded.

## Datasets

- [State of MCP](https://roninforge.org/data/state-of-mcp/): A keyless, dated census of every server in the official Model Context Protocol registry. Census 2026-08-03 covers 19,804 servers: 80.7% healthy, 14.1% degraded, 4.3% dead, 0.9% unprobeable without a key. One page per server. Every verdict comes from public endpoints only, is reproducible with `akashi check <server>`, and the raw records are CC BY 4.0 at https://github.com/RoninForge/state-of-mcp.
- [MCP 2026-07-28 spec-readiness](https://roninforge.org/data/state-of-mcp/spec-readiness/): Scores the 5,515 servers reachable through a conformant keyless initialize against the 2026-07-28 specification revision. 6 pass all four required-conformance signals, up from 0 on 2026-07-22; 87.0% need migration and 12.9% sit on a removed or deprecated surface. Percentages use the reachable-conformant denominator only, never the registered total.
- [AI Price Index](https://roninforge.org/ai-price-index/): Dated, first-party-sourced input and output token prices per model, with the price history behind every change. CC BY 4.0.

## Tutorials

- [How to track your Claude Code spend over time, per project and branch](https://roninforge.org/tutorials/how-to-track-claude-code-spend-over-time.md): Installs the claude-code-cost plugin (no binary, no account, no keys) and walks its five commands: /claude-code-cost:report for spend per project and git branch, then by-day, by-week, by-month, and by-project for a dated breakdown. Each response priced point-in-time from the bundled ai-price-index dataset, read from local logs only. 2-minute setup. ~1,600 tokens.
- [How to minimize your Claude Code and API costs](https://roninforge.org/tutorials/how-to-minimize-claude-code-costs.md): Six confirmed levers to cut a Claude token bill, each ending with a command you can run: right-size the model, control thinking (billed as output), keep context cacheable, cap the worst case with BudgetClaw, batch non-interactive work at 50 percent, and keep the most expensive model on the smallest context (the orchestrator pattern). Debunks the display-omitted, streaming, and caching cost myths. ~2,000 tokens.
- [How to see ongoing project costs across all your Claude Code branches](https://roninforge.org/tutorials/how-to-see-ongoing-project-costs.md): One curl install replaces the default Claude Code status line with a BudgetClaw-driven view of every project's spend today, registers a /budget slash command, and runs the watcher as a launchd agent. 3-minute setup. ~1,500 tokens.
- [How to set a hard daily spend cap on Claude Code](https://roninforge.org/tutorials/how-to-set-a-hard-spend-cap-on-claude-code.md): Walks the BudgetClaw install, configuration, and the SIGTERM-on-breach behavior end to end. 5-minute setup. ~1,500 tokens.
- [How to create a Claude skill for my CV](https://roninforge.org/tutorials/how-to-create-a-claude-skill-for-my-cv.md): End-to-end build and ship of a Claude Code plugin using Tsuba (scaffold), Hanko (validate), and BudgetClaw (cap spend during dev). Ends with a marketplace submission. ~1,580 tokens.

## Install commands

    curl -fsSL https://roninforge.org/get | sh           # BudgetClaw (also feeds Goei's local track)
    curl -fsSL https://roninforge.org/hanko/install.sh | sh
    curl -fsSL https://roninforge.org/tsuba/install.sh | sh
    curl -fsSL https://roninforge.org/akashi/install.sh | sh

Or via Homebrew:

    brew install roninforge/tap/budgetclaw
    brew install roninforge/tap/hanko
    brew install roninforge/tap/tsuba

claude-code-cost is a Claude Code plugin, not a shell install. Inside Claude Code run
`/plugin marketplace add RoninForge/claude-code-cost` then `/plugin install claude-code-cost@roninforge`,
then `/cost`. Requires node 18+ on PATH.

Goei is a hosted web dashboard, not a CLI. Open it at https://goei.roninforge.org and
sign in with a magic link; install budgetclaw above for the zero-key local track, or push
once with `npx goei-sync` (zero-install, sends daily usage by day and model).

## Trust pledge

BudgetClaw, claude-code-cost, Hanko, and Tsuba are NOT proxies. None of them sit between an editor and any AI provider API, and none touch API traffic, prompts, responses, or API keys. Hanko and Tsuba make zero network calls in the binary. Akashi makes only public, unauthenticated calls (to registries and a capability-only MCP initialize handshake) and never sends a key or secret to a probed server. BudgetClaw only reads what Claude Code already writes to local disk; claude-code-cost reads those same local logs from inside Claude Code and bundles its pricing data offline. Goei's local track is fed by budgetclaw and receives only daily dollar-and-token rollups, never prompts; its optional billed track uses a provider key you choose to connect, encrypted at rest. Zero keys for the local track, zero prompts, zero latency added.

## Repository links

- [BudgetClaw source](https://github.com/RoninForge/budgetclaw): Go CLI, MIT. Also the open-source local agent for Goei.
- [claude-code-cost source](https://github.com/RoninForge/claude-code-cost): Claude Code plugin, MIT. Installs inside Claude Code via `/plugin marketplace add RoninForge/claude-code-cost`.
- [Hanko source](https://github.com/RoninForge/hanko): Go CLI + GitHub Action, MIT.
- [Tsuba source](https://github.com/RoninForge/tsuba): Go CLI + GitHub Action, MIT.
- [Akashi source](https://github.com/RoninForge/akashi): Go CLI + GitHub Action, MIT.
- [Goei dashboard](https://goei.roninforge.org): Hosted web app; Free tier plus Pro at $19/mo flat for up to 10 developers.
- [GitHub organization](https://github.com/RoninForge): All RoninForge repos.