Agentero: CLI and MCP server for a research library
Agentero
A CLI and MCP server for a research paper library: the agent searches, imports and reads papers and keeps Markdown notes in your vault
- 1,117 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Through `file_write`, `paper_notes_write` and the move and delete tools the agent can change and delete library files
- The MCP port needs no authentication: any local process gets access to the library
- With `--expose-paper-text` the full PDF text goes to the client
Install
- Works in
Manual install
brew install agenteroInstalls the CLI. Run `brew tap poco-ai/agentero` first; the app installs as `brew install --cask agentero`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Through `file_write`, `paper_notes_write` and the move and delete tools the agent can change and delete library files
- The MCP port needs no authentication: any local process gets access to the library
- With `--expose-paper-text` the full PDF text goes to the client
This is third-party code. Review the repository files before installing.
What it does
Agentero is a desktop app for research work: reading PDFs with annotations, Markdown notes with backlinks, translation and a paper library. The library lives on disk as plain files (PDF and Markdown). For agents the project provides the `agentero` CLI and a built-in MCP server through which Claude Code, Codex and other clients work with that library. The MCP server listens on loopback only, at `http://127.0.0.1:8765/mcp`. You turn it on in the app settings (Settings, General, MCP server) or run it without a window with `agentero mcp serve --vault <path> --listen 127.0.0.1:8765`. Tools: `paper_list` and `paper_get` return metadata, `import_id` imports by arXiv number, DOI or URL, `paper_notes_get` and `paper_notes_write` work with a paper's notes file, and there are tags and search across all Markdown files in the vault (`vault_search`). Reading and writing text files outside the `papers` folder goes through `file_list`, `file_read` and `file_write`. The resources `agentero://vault`, `agentero://agent-invariants` and `agentero://skills/agentero-cli` explain the working rules to the agent. Separately, the app connects local agents over the ACP protocol and can expose the library to ChatGPT through an OpenAI tunnel.
Who it is for. For researchers and graduate students who keep a paper library and want Claude Code or Codex to search it, import papers and write notes.
Good fit when
- You want the agent to put a paper into the library by arXiv number or DOI and create a note for it
- You are looking for a mention of a method in your notes and ask the agent to search the whole vault
- The library sits on a remote server and the agent must work with it without the desktop app
- You want to tag papers and update metadata (journal, date, DOI) in bulk
- You keep a paper draft in the vault and let the agent read and edit text files next to the literature
Not a fit when
- You need a shared library for a group: the vault is local and syncs through S3-compatible cloud storage
- You need access to a remote server over the network: the MCP port has no authentication and is meant for loopback only
- Your library lives in Zotero and you do not want to change habits: an MCP server for Zotero fits better
- You need to give the AI full PDF text by default: it is closed until a separate flag is enabled
Example request
Import the paper with DOI 10.1145/3292500.3330701 into the library, add the tag baseline and write three key conclusions into its note.Limitations
The server has no authentication and is bound to loopback, so the port must not be exposed to the network, and access from another machine goes through an SSH tunnel. In windowless mode you must name the vault explicitly; for the app, a vault must be open and the program running. PDF text is not given to clients until `--expose-paper-text` is enabled. Remote vaults are not served over MCP.
Connecting to ChatGPT requires an OpenAI key with tunnel permissions and the `tunnel-client` program. Linux needs Ubuntu 22.04 or newer. The main README and docs are in Chinese, and an English README exists. Availability of the apps from Russia was not checked for this card.
How to disable. In the app turn off the MCP server switch in settings; for windowless mode stop the `agentero mcp serve` process. Then remove the server entry from your client's MCP settings.
FAQ
How do I connect Agentero to Claude Code?
Turn on MCP server in the app settings or run `agentero mcp serve --vault ~/papers`. The server listens at `http://127.0.0.1:8765/mcp`; connecting clients is covered in the project docs: https://github.com/poco-ai/agentero/blob/HEAD/docs/usage/mcp.md.
How do I install the app and the CLI?
On macOS through Homebrew: `brew tap poco-ai/agentero`, then `brew install --cask agentero` for the app or `brew install agentero` for the CLI. Installers for all systems are on the releases page.
Does the app have to stay open?
For the app's server, yes: it runs only while the app is running. Without a window, `agentero mcp serve` does the job: the vault is set with the `--vault` flag or the `AGENTERO_VAULT` variable.
What can the agent change in the library?
Paper metadata, tags and notes; it can import new papers and write text files outside the `papers` folder, for example a draft `drafts/main.tex`. Notes are written with `paper_notes_write`, while `file_write` cannot touch PDFs, the `.agentero` service folder or the notes themselves. The server asks the agent to confirm before overwriting a note.
How is Agentero different from Zotero?
Agentero keeps papers and notes as plain files and hands them to agents through MCP, CLI and ACP. You can import a whole Zotero database with tags, notes and attachments, and the Zotero browser extension keeps saving papers.
Discussion
No questions or reviews yet.
Related
CLI toolsA local MCP server for arXiv literature work: section-level paper reads, BibTeX from metadata, topic watches
A local MCP server lets an agent search your Zotero library, Obsidian vault and Apple Notes through a C++ search core
A Zotero plugin with a built-in MCP server: the agent searches the library, reads PDFs and notes, and works with annotations and metadata
Two skills for Claude Code and Codex: the agent translates papers in the EasyRead library and answers your margin notes
A PDF-reading MCP server: structured text, tables, OCR, visual crops and page-level citations
MCP for PDF and ebook files: the agent finds books in a folder and reads EPUB and PDF by chapter and page, with contents and metadata as Markdown
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Agentero: CLI and MCP server for a research library (this card) | no | 1,117 | ||
| Zotero MCP | no | key is optional | 5,764 | connects an existing Zotero library; Agentero keeps its own vault of plain files |
| arxiv-mcp-server | no | not required | 3,258 | searches and reads arXiv papers; Agentero keeps your library and notes locally |
README in short
The README explains the idea: a paper library, notes and model conversations are usually scattered across programs, and two-column PDFs are hard for models. So the app turns formulas, tables and algorithms into structured Markdown and LaTeX. It continues with translation using your own key (BYOK), arXiv recommendations, reference-list parsing, S3 sync and migration from Zotero.