Librarium CLI
A CLI and MCP server that sends one question to search and AI sources and saves answers, citations and cost in a run.
- 137 ★ on the repository
- Last commit
Extended access
Why this level
- Questions and cited pages go to the providers you pick, and calls spend money on your keys.
- Runs are saved to `./agents/librarium` and can contain private data from your questions.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/librariumDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Questions and cited pages go to the providers you pick, and calls spend money on your keys.
- Runs are saved to `./agents/librarium` and can contain private data from your questions.
This is third-party code. Review the repository files before installing.
What it does
Librarium sends one research question to the sources you choose and saves each answer, citation and cost in a resumable Run. There are sixteen sources: Brave, Claude, Exa, Firecrawl, Gemini, Google, Grok, Jina, Kagi, OpenAI, OpenRouter, Parallel, Perplexity, Tavily, Valyu and You. Five more engines record what a consumer AI product showed (ChatGPT, Gemini App, Google AI Mode, Google AI Overview, Copilot) through SearchAPI. `librarium mcp` runs a stdio MCP server with three tools: `research`, `get_run` and `list_sources`. The `--deep` mode submits long research tasks and `show` fetches the result later without resubmitting. `librarium install-skill` puts `SKILL.md` into `~/.claude/skills/librarium`.
Who it is for. Researchers and developers who compare answers from several search and AI services and want sources and spend per service.
Good fit when
- Managed PostgreSQL services are compared and the answer must come from several sources with citations.
- Deep research is launched with `--deep --no-wait` and collected later with `show`.
- The budget is tight: the run uses `--max-cost 1` and sources beyond the budget are skipped.
- How ChatGPT, Gemini and Google AI Mode describe a brand is checked through a saved engine group.
Not a fit when
- You hold no key for any source: at least one is needed, for example `EXA_API_KEY`.
- The question contains private data: it goes to providers while answers and citations are saved to disk.
- Spending must never exceed the cap by a cent: calls already in flight can finish above the limit.
Example request
Compare managed PostgreSQL services through Exa and OpenAI, cap spend at one dollar and show the citations.Limitations
It needs Node.js 22.12 or newer (Homebrew and standalone binaries bring their own runtime) and an API key for at least one source; provider calls cost money. A cost cap skips sources without a price estimate, and calls already started can finish above the limit. Cost comes only from the provider, unknown cost is stored as `null`, and a failed call may still be billed. Provider answers and citations count as untrusted data. Version 2.0 migrates a 1.x config to format 3 with a backup, and `show` cannot read 1.x run directories.
How to disable. Run `claude mcp remove librarium`, delete `~/.claude/skills/librarium` and uninstall the package with `npm uninstall -g librarium` or `brew uninstall librarium`. Saved runs live in `./agents/librarium` and are deleted separately.
FAQ
How do I test a run without spending money?
Add the `--dry-run` flag: `librarium research "question" --only exa --dry-run`. The command plans the request, calls no provider and saves no run.
How do I connect Librarium to Claude Code?
Run `claude mcp add --scope user librarium -- librarium mcp`. The skill installs with `librarium install-skill`. Steps for other agents are in the guide at librarium.si.
What does deep mode mean?
The `--deep` flag submits long tasks to providers. With `--no-wait` the command returns at once, and `librarium show <run-id> --wait` collects the result later without resubmitting.
How do sources differ from engines?
Sources perform the research. Engines record what a consumer AI product showed, run only when named and mark results as `observation: "surface"`.
Which keys do I need?
One source key is enough, for example `EXA_API_KEY`. The `librarium setup` wizard accepts literal values or references like `env:EXA_API_KEY`, and environment variables work without a config file.
Discussion
No questions or reviews yet.
Related
CLI toolsGPT Researcher MCP
GPT Researcher MCP Server
An MCP server for deep web research: GPT Researcher vets sources on its own and hands the agent a ready report
A Claude Code skill that turns the agent into a researcher: multi-step source collection, citation verification, and a wiki report that grows session by session
Linkup MCP server: real-time web search, page fetching and long research tasks with sources for an agent
A local research MCP server: web, Reddit and YouTube search, maps and code exploration for a model in LM Studio or Cursor
An MCP server that searches Google and Scholar, reads pages and PDFs and stores findings in a local project knowledge graph
A unified search MCP server: one interface to Tavily, Brave, Kagi, Exa, Linkup, Firecrawl and GitHub search
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Librarium (this card) | no | 137 | ||
| Exa MCP Server | yes | OAuth | 5,213 | the official server of a single search vendor; Librarium queries sixteen sources and tracks cost |
| GPT Researcher MCP | yes | API key | 376 | one deep research by the GPT Researcher agent; here one question goes to several providers with the run saved |
README in short
The README moves from installation and a first run to deep mode, source choice, MCP, the JavaScript API and limits. Cost is described separately: it comes only from the provider's report.