nodebench — Claude Code plugin: features and installation
nodebench is a Claude Code plugin with MCP tools for company research, competitor comparisons, and code QA. It reveals extra toolsets as tasks call for them.
MIT ★ 14 updated 2026-10-23
/plugin marketplace add anthropics/claude-plugins-community
/plugin install nodebench@claude-communityWhat nodebench does
nodebench gives Claude Code MCP tools for research, comparison, search, and reporting. Its default set has 9 tools, including investigate, compare, and report. Use discover_tools to find more, then load_toolset to add a toolset for a specific task.
The plugin adds /nodebench-qa for a full QA run and /agentic-reliability-audit for an infrastructure check. Its rules trigger QA after code changes and include a 5-whys diagnostic method plus an agent reliability checklist. Manual MCP setup in a project uses npx and the nodebench-mcp package.
What it helps with
- create a sourced company research report
- compare competitors in a side-by-side brief
- find tools for crawling and inspecting a site
- run QA after a code change
Example prompts for the agent
- Research this company and prepare a report with sources.
- Compare these competitors in a side-by-side brief.
- Run the full QA process with
/nodebench-qa.
How to install nodebench
Plugin marketplace
/plugin marketplace add anthropics/claude-plugins-community/plugin install nodebench@claude-communityStep by step, with commands for each platform: Claude Code plugins: marketplaces, installing, and building your own
Questions about nodebench
How do I install nodebench in Claude Code?
Run these commands in order: /plugin marketplace add anthropics/claude-plugins-community, then /plugin install nodebench@claude-community.
What does nodebench do?
It adds MCP tools for research, comparison, search, and reports, along with QA and agent infrastructure audit commands.
Which MCP tools are available by default?
The default set contains 9 tools, including investigate, compare, track, summarize, search, and report. Find extra sets with discover_tools and add one using load_toolset.
Source text
claude-plugin · MIT
# NodeBench - Claude Code Plugin
NodeBench MCP inside Claude Code. Start with the same 9-tool workflow lane the main package exposes, then load deeper toolsets only when the task actually needs them.
## What you get
- **MCP Server**: 9-tool default lane for `investigate`, `compare`, `track`, `summarize`, `search`, `report`, `ask_context`, `discover_tools`, and `load_toolset`
- **Slash Commands**: `/nodebench-qa` (full QA loop), `/agentic-reliability-audit` (8-point infra check)
- **Rules**: Auto-QA after code changes, analyst diagnostic (5-whys), agentic reliability checklist
## Install
```bash
/plugin install nodebench@claude-plugins-official
```
Or add to your project's `.mcp.json`:
```json
{
"mcpServers": {
"nodebench": {
"command": "npx",
"args": ["-y", "nodebench-mcp", "--preset", "default"],
"env": {}
}
}
}
```
## Quick start
1. `investigate({ topic: 'analyze a company' })` - get a sourced report first
2. `compare({ items: ['Competitor A', 'Competitor B'] })` - produce a side-by-side brief
3. `discover_tools({ query: 'crawl and inspect a site' })` - find the deeper lane only if needed
4. `load_toolset({ toolset: 'site_map' })` - expand deliberately when the task calls for it
## Links
- [GitHub](https://github.com/HomenShum/nodebench-ai)
- [Website](https://nodebenchai.com)
- [npm](https://www.npmjs.com/package/nodebench-mcp)
© HomenShum · MIT · HomenShum/NodeBenchAI