frontier MCP server
Commit Nov 14, 2026Searches arXiv, GitHub, and Hugging Face to provide live research context, papers, repositories, and models to AI clients.
Also known as revoydotdev/frontier-mcp
- Stars
- 0
- Tools
- 9
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 14, 2026
- License
- MIT
What is frontier MCP?
frontier searches arXiv, GitHub, and Hugging Face to provide live research context, papers, repositories, and models to AI clients. It exposes 9 tools, such as survey, arxiv_search, arxiv_get.
Summary generated from the project's README
frontier MCP tools 9
Detected from the READMEsurveyarxiv_searcharxiv_getarxiv_fetchgh_search_reposgh_repohf_search_modelshf_search_datasetshf_model
Automated checks
Ran Nov 15, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablerevoydotdev/frontier-mcp is public and reachablegithub.com/revoydotdev/frontier-mcp ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-04github.com/revoydotdev/frontier-mcp ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model9 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubfrontier-mcp
Live research context for MCP clients
Search primary material across arXiv, GitHub, and Hugging Face without leaving the agent workflow.
Quick start · Tools · Operational boundaries · Development
frontier-mcp is a stdio Model Context Protocol server for research that benefits from current, attributable material. It searches papers, implementation repositories, models, and datasets, then returns their identifiers, links, metadata, and selected primary content in a stable result envelope.
The server is intentionally narrow: it is a retrieval surface, not a recommender or an execution environment. Start with a cross-source survey, follow the source records that matter, and cite the paper ID, repository, or model/dataset ID in the resulting work.
Quick start
Requirements
- Python 3.11 or newer
- uv (recommended for an isolated, locked install)
- Network access to the sources you use
Clone the repository and create the project environment:
git clone https://github.com/revoydotdev/frontier-mcp.git
cd frontier-mcp
uv sync --locked
Add the server to Claude Code at user scope:
claude mcp add -s user frontier -- "$PWD/.venv/bin/python" -m frontier_mcp
Any MCP client that supports stdio can run the same command and arguments:
command: /absolute/path/to/frontier-mcp/.venv/bin/python
args: ["-m", "frontier_mcp"]
The command starts a stdio server; it is not an interactive CLI. Use your MCP client's tool inspector or a connected agent to call it.
First query
Ask the client to call:
survey("retrieval-augmented generation")
The response combines the top matching arXiv papers, GitHub repositories, and Hugging Face models. Use its hits counts to see which sources contributed, then use the provider-specific tools to inspect promising records.
Tools
Every tool returns one of these envelopes, so clients can handle provider failures consistently:
{"ok": true, "result": {}}
{"ok": false, "error_code": "ExceptionType", "error_detail": "…"}
| Tool | Purpose | Key parameters |
|---|---|---|
survey |
Search papers, repositories, and models together. A failure from one provider is reported in that provider's field without discarding the other results. | topic, max_each=5 |
arxiv_search |
Search arXiv paper records. Results include title, authors, abstract, categories, dates, and paper/PDF URLs. | query, max_results=10, category, sort=relevance|submitted|updated |
arxiv_get |
Retrieve a full arXiv record by identifier. | arxiv_id |
arxiv_fetch |
Download a paper PDF for local reading and return its filesystem path. | arxiv_id, out_dir |
gh_search_repos |
Search GitHub repositories with maturity and licensing metadata. | query, language, sort=stars|updated, limit=10 |
gh_repo |
Retrieve repository metadata and, when GitHub returns it, a README excerpt capped at 5,000 characters. | repo (owner/name) |
hf_search_models |
Search Hugging Face models. | query, task, sort=downloads|likes|modified|created|trending, limit=10 |
hf_search_datasets |
Search Hugging Face datasets. | query, sort=downloads|likes|modified|created|trending, limit=10 |
hf_model |
Retrieve model-card metadata, including available license, base-model, and dataset fields. | model_id |
For repository recommendations, preserve the returned stars, forks, last-push time, and license alongside the repository URL. Those details help distinguish active, broadly adopted code from a merely relevant search hit.
Authentication and local files
| Source | Credentials | Behavior |
|---|---|---|
| arXiv | None | Paper search and retrieval work without a key. |
| GitHub | GITHUB_TOKEN or GH_TOKEN; otherwise gh auth token when available |
Authentication raises rate limits and may expose repositories the token can access. Public search works without it, subject to GitHub's unauthenticated limits. |
| Hugging Face | Cached Hugging Face token when present | Public model and dataset search works without a token. |
All network operations are read-only: this server does not create, modify, or submit content to arXiv, GitHub, or Hugging Face. arxiv_fetch is the one local write: it saves a PDF under ~/frontier-material/arxiv by default, or in the supplied out_dir.
Operational boundaries
The three sources are deliberately useful but incomplete. They favor academic and ML-oriented material; systems, kernel, and vendor-tool research may instead require official documentation, changelogs, issue trackers, or other web sources. A zero-result survey is useful evidence about the query or source coverage, not proof that the subject does not exist.
Search output is current at retrieval time, not a quality, security, compatibility, or licensing determination. Inspect the linked primary material before adopting a dependency or making a claim. In particular, repository metadata is a signal to cite and assess, not an endorsement.
Development
The locked development environment and the repository's CI use:
uv sync --extra dev --locked
uv run ruff check .
uv run pytest
Run the server directly during local MCP-client development:
uv run frontier-mcp
The test suite covers the stable error envelope and survey fan-out behavior, including partial provider failures and the all-zero-results guidance path.
License
MIT © 2026 revelri
frontier MCP questions
What tools does frontier MCP provide?
It exposes 9 tools, including survey, arxiv_search, arxiv_get, arxiv_fetch, gh_search_repos.
Can the server modify my GitHub or arXiv accounts?
No, all network operations are read-only and the server cannot create, modify, or submit content to these platforms.
How can I download an arXiv paper?
You can use the arxiv_fetch tool to download a paper PDF to a local directory.