doiget
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@doigetget the PDF for arXiv:2103.12345"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
doiget
A single-binary CLI + stdio MCP server that turns DOIs and arXiv ids into local PDFs through official, OA-first APIs. Designed as the agent-facing companion to BiblioFetch.jl.
Docs: stable = the Zola site (built from main); dev = rustdoc built from next; API = docs.rs (latest published release).
Status: Shipping on crates.io (doiget-core, doiget-cli,
doiget-mcp), with sigstore-signed binaries + an SBOM attached to the GitHub
Release. Tier 1 + Tier 2 sources, the stdio MCP server, citation-graph
expansion, and gated TDM sources are all implemented. Releases are cut by a
single signed git tag through the tag-driven pipeline (see
ADR-0025); release-plz was
retired. See CHANGELOG.md for history and
docs/PHASES.md for the phase plan.
Posture
doiget is a general-purpose automation tool for retrieving papers via official publisher APIs. By design, doiget only attempts retrieval through:
Public Open Access sources (default — Crossref, Unpaywall, arXiv).
Credentials the user has personally configured for their own institutional or personal subscriptions (opt-in, compile-time gated).
doiget does not work around any access control mechanism, redistribute papers, host content, operate as a SaaS, or bundle any publisher API keys.
Users are responsible for ensuring they have the right to access the content they request and for compliance with each source's Terms of Service.
See docs/LEGAL.md and docs/SCOPE.md.
Related MCP server: Paper Search MCP
Documentation
Reader | Entry point |
CLI user | This README, then |
Agent / MCP host integrator | |
Library user (Rust) | docs/PUBLIC_API.md + crates.io rustdoc |
Contributor | |
Publisher legal team | |
Security researcher | docs/SECURITY.md + docs/PROVENANCE_LOG.md + docs/CAPABILITY.md |
BiblioFetch.jl user |
Architecture: docs/ARCHITECTURE.md Threat model: docs/SECURITY.md Permanent non-goals: docs/SCOPE.md Phase plan: docs/PHASES.md ADRs: docs/DECISIONS/
Installation
doiget ships a single self-contained binary — the Linux build is fully static (musl), so it runs on old glibc / HPC boxes too. Every channel installs the same checksum-verified binary from the signed GitHub Release.
Recommended: use a prebuilt binary (no Rust toolchain, no compiler). The shell / PowerShell installers below download the signed release binary directly.
cargo install(further down) instead compiles from source and therefore needs a working C/C++ build toolchain — see that section for the per-platform requirements.
Shell installer (Linux / macOS)
curl -fsSL https://raw.githubusercontent.com/QAtlasHub/doiget/main/scripts/install.sh | shInstalls to ~/.local/bin (override with DOIGET_INSTALL_DIR); pin a version with
DOIGET_VERSION=0.6.0. The script verifies the published SHA-256 sidecar before installing.
PowerShell installer (Windows)
irm https://raw.githubusercontent.com/QAtlasHub/doiget/main/scripts/install.ps1 | iexFrom crates.io (Rust toolchain — compiles from source)
The published crate is doiget-cli (it produces the doiget binary).
cargo install doiget does not work — there is no crate by that bare name:
cargo install doiget-cli # installs the `doiget` binaryBecause this compiles from source, you need a C/C++ build toolchain (a
linker is mandatory — cargo install cannot link a binary without one):
Linux:
gcc/clang+make(e.g.build-essential).macOS: Xcode Command Line Tools (
xcode-select --install).Windows: either the Visual Studio Build Tools with the "Desktop development with C++" workload (provides
link.exe), or a MinGW-w64 toolchain used via the GNU target (rustup toolchain install stable-x86_64-pc-windows-gnu+cargo +stable-x86_64-pc-windows-gnu install doiget-cli).
If you don't have (or don't want) a build toolchain, use one of the prebuilt installers above — they need no compiler.
npm / npx — one line in an agent config
npx -y doiget-cli serve # MCP server, no install step
npm install -g doiget-cli # or put `doiget` on PATHThe package is doiget-cli and the command it installs is doiget — the same
shape as cargo install doiget-cli, which is where the name comes from. npm
refuses the unscoped name doiget as too similar to the unrelated giget, and
matching the crate turned out to be the better answer regardless: one name for
the tool on both registries.
The npm packages carry the same signed release binaries as optionalDependencies
— npm resolves the one matching your platform, and there is no postinstall
download, so this works under --ignore-scripts and through a corporate
registry mirror. npm view doiget version tells you what is published; the
packages ship from tagged releases, so a very new commit may be ahead of them.
Homebrew
brew tap QAtlasHub/doiget https://github.com/QAtlasHub/doiget
brew install doigetThe tap lives in this repository rather than a separate homebrew-doiget, which
is why the tap line carries an explicit URL. A dedicated tap repo would shorten
it to brew tap QAtlasHub/doiget; the formula would move across unchanged.
Formula/doiget.rb installs the same signed release binary the GitHub
Release publishes, pinned by the sha256 from that release's own .sha256
asset — the file the shell installer verifies against, so the two channels
cannot disagree about what they installed. It is generated by
scripts/update-homebrew-formula.sh, never hand-edited, and CI fails if the
committed formula is not what the generator produces.
The formula tracks the latest stable release, so it can trail a very new
tag by one commit; brew info doiget shows which version it pins.
Claude Code plugin
/plugin marketplace add QAtlasHub/doiget
/plugin install doiget@doigetReads .claude-plugin/ from this repository's default branch. The plugin's
.mcp.json runs npx -y doiget-cli serve, so it needs nothing installed
beforehand — npm fetches the wrapper and the one matching platform binary on
first run. Until this it ran a bare doiget, which meant the plugin worked
only for people who had already installed doiget some other way; that is why it
was self-hosted rather than submitted to the Anthropic plugin directory, and
why it is now submittable.
Channel status
Channel | Status |
Shell / PowerShell installer | shipping |
GitHub Release binaries (signed, SBOM) | shipping |
| shipping (needs a C linker) |
| shipping since 0.8.4 |
MCP Registry | listed |
npm / npx |
|
Claude Code plugin | self-hosted marketplace, as above |
Homebrew |
|
Nix |
|
| not built — low value; most Linux users take the binary or Nix |
Docker | not planned — see below |
Docker was ranked second in #501 on the grounds that a container is "the only
architecture the Tier-3 features can legally be used in". That premise does not
hold: ADR-0002 decides that the default published binary contains no TDM
source code at all, and an image built from the published binary would ship
oa-only,citation like every other prebuilt channel. The architecture the
sentence describes is real, and it is served by building from source with
--features tdm-<publisher>, which is not a distribution channel. What remains
is "a shape enterprises can pin and scan" — worth something, but doiget is a
single statically-linked binary, so a container solves no dependency problem
here and the existing .sha256 plus cosign bundle already give a pinnable,
verifiable artefact.
npm was the one channel whose pipeline was written and whose packages did not
exist. npm Trusted Publishing cannot perform a package's first publish — the
setting lives under a package's Settings, and there is no Settings page for a
package that has never been published — so the release job, which carries no
token by design, cannot create them. scripts/bootstrap-npm.sh does the
once-only placeholder publish that unblocks it; CONTRIBUTING.md has the
runbook.
As of 2026-08-27 the four per-platform packages are published as 0.0.0
placeholders. npm points latest at a package's first publish whatever
--tag says, so those placeholders are deprecated — that notice is the only
warning until a release moves latest to a real version.
#247 was closed as completed
while four of its five channels did not exist; the remaining ones are tracked in
#501. Every release asset is
cosign-keyless signed (<asset>.cosign.bundle) for optional verification.
Quick start
# Fetch a paper by DOI (Open Access only by default)
doiget fetch 10.1103/PhysRevLett.130.200601
# Fetch by arXiv id
doiget fetch arXiv:2401.12345
# Batch fetch
doiget batch refs.txt
# Verify a bibliography's references resolve (no PDF download) — CI gate
doiget verify docs/references.bib --strict
# Lint a .bib for structural issues (no network): missing fields,
# blank fields, $$-display-math titles. Read-only and math-aware.
doiget lint docs/references.bib
# Inspect what was fetched
doiget info 10.1103/PhysRevLett.130.200601
# Run as MCP server (stdio)
doiget serveDefault features fetch only Open Access PDFs through Crossref / Unpaywall / arXiv. Institutional TDM access (Springer OA, APS Harvest, Elsevier ScienceDirect TDM) is not in the default published binary; it must be opted in at build time per publisher. See docs/SOURCES.md.
Coexistence with BiblioFetch.jl
doiget and BiblioFetch.jl share the same on-disk store format (TOML metadata + PDF files
under a configurable store root). doiget defaults to ./papers (under the current working
directory; ADR-0036), BiblioFetch.jl to ~/papers/; point both at the same root (e.g.
DOIGET_STORE_ROOT=~/papers) to share one store. The shared schema, locking protocol, and
atomic write contract are
specified in docs/STORE.md. Reference test vectors for the shared safekey
algorithm are in docs/SAFEKEY.md.
Tool | Best for |
BiblioFetch.jl | Julia REPL, research vault, citation graph exploration |
doiget | Agents / MCP hosts, batch operations, scripted pipelines, container deployments |
License
MIT for the doiget source code and binaries (see LICENSE).
The license under which doiget retrieves papers is separate and is determined by each paper's own license, the publisher's API Terms of Service, and the user's own access rights. doiget does not relicense fetched content.
Contact
Takedown requests, security disclosures, and other formal correspondence: CONTACT.md.
This server cannot be deployed
Maintenance
Related MCP Connectors
Search arXiv/Semantic Scholar/OpenAlex + medical evidence (PubMed/Europe PMC) + LaTeX/PDF tools.
Generate and read PDFs for AI agents: a generate_pdf and a read_pdf tool, priced per document.
Search and download academic papers from arXiv, PubMed, bioRxiv, medRxiv, Google Scholar, Semantic…
Find academic papers across major sources like arXiv, PubMed, bioRxiv, and more. Download PDFs whe…
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables searching, downloading, and managing academic papers from arXiv.org through natural language interactions. Provides tools for paper discovery, PDF downloads, and local paper collection management.41MIT
- AlicenseNot gradedqualityDmaintenanceEnables searching and downloading academic papers from multiple sources including arXiv, PubMed, bioRxiv, Google Scholar, and Semantic Scholar. Provides standardized tools compatible with OpenAI Deep Research and ChatGPT connectors.15MIT
- AlicenseAqualityDmaintenanceEnables LLMs to search, download, and read arXiv papers with automatic PDF text extraction and section filtering. Provides AI assistants direct access to scientific literature with local caching for fast re-access.31MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI agents to download, index, and semantically search PDF research papers using 8 MCP tools.2GPL 3.0