Best Claude Tools
CLI tools, editors, and utilities built for Claude.
Standalone Tools for Claude
Tools are standalone CLIs, wrappers and utilities for the Claude ecosystem, projects that don't fit cleanly as a skill, MCP server or plugin but still belong in the orbit. Think: cost calculators, prompt managers, transcript viewers, conversation exporters, model routers, evaluation harnesses.
These projects often power workflows that sit alongside Claude Code rather than inside it. A common pattern: a CLI that batches prompts to the Claude API, parses results, and writes a CSV, useful for evals, content pipelines, and anything that needs deterministic batch processing.
Below: 790+ tools ranked by stars and Trust Score. Filter by use case to narrow (data, devops, evaluation, content…).
Questions about Tools
How is a 'tool' different from an MCP server?+
An MCP server exposes capabilities to Claude over the protocol, Claude calls it. A standalone tool runs independently and may call Claude (or not). MCP servers are reactive; tools are proactive. Many projects ship as both, a CLI for humans plus an MCP server for Claude.
What kind of tools live in this category?+
Cost trackers (claude-cost), prompt CLIs (claude-cli, llm), evaluation harnesses, transcript exporters, model gateways (LiteLLM, OpenRouter wrappers), and project setup utilities. Anything that's not a skill/agent/MCP/plugin/template/hook.
Are these tools cross-model (work with OpenAI too)?+
Many are. Tools that wrap the Anthropic SDK only work with Claude; tools built on LiteLLM or LangChain work across providers. Each repo's README states the supported providers, check before adopting.
Do tools need API keys?+
Most do. The Anthropic ones need ANTHROPIC_API_KEY; cross-provider ones need keys for whichever providers you enable. Keys are usually read from environment variables, never paste them into config files committed to git.
Can I use Claude tools in my CI/CD pipeline?+
Yes. Many tools in this category are explicitly built for CI, they accept env vars, write structured output (JSON/CSV), and exit non-zero on failure. Common patterns: PR review bot, doc generation step, accessibility check, security scan.