Guide · Agent skill
How to use the BGPT Paper Search agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The BGPT Paper Search skill teaches an AI agent how to search the BGPT scientific paper service by topic or DOI and return structured evidence: experiments, reported statistics, scope, and limitations pulled from full papers, not just titles and abstracts. This guide is for researchers, students and analysts who want that kind of detail. By the end you will know what the skill does, how to install it, how to connect the BGPT server it depends on, and how to phrase requests that trigger it.
BGPT Paper SearchK-Dense-AI/scientific-agent-skills · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the skill ZIP from this page and unzip it so the bgpt-paper-search folder with SKILL.md is in ~/.claude/skills/ or your project's .claude/skills/ folder.
- 2Or upload it in the Claude apps. In the Claude apps, open Settings, go to Capabilities, then Skills, and upload the skill.
- 3Configure the BGPT MCP server. Add the BGPT remote endpoint (Streamable HTTP or SSE) to your host app, or use the mcp-remote bridge if your host needs a stdio config.
- 4Add an API key for paid use. Free access needs no key. For paid usage, supply your key as an Authorization Bearer header through the host's credential storage.
- 5Ask a research question. Describe the topic or give a DOI in plain language. The agent will search or look up the paper using the connected tools.
- 6Verify the evidence. Check the DOI, extraction status and warnings, then confirm sample sizes and statistics against the original paper.
What the BGPT Paper Search skill does
An agent skill is a folder with a SKILL.md file that gives an AI agent instructions for a specific task. This one covers literature search. It tells the agent how to query the BGPT scientific paper service and how to read what comes back.
The skill is built around evidence records. Each record links a central claim to an experiment, reported statistics, the demonstrated scope, and provenance. The skill also tells the agent to keep track of DOIs, query history, and truncation or legacy-record warnings so findings can be traced back to the source.
- Search by topic or search for a known paper by DOI
- Returns extracted claims with their supporting experiments and statistics
- Flags records that are truncated or in an older format
- Keeps DOIs and query history so you can trace each finding
Install the skill (ZIP or Claude app upload)
Download the skill ZIP from the skill's page on this directory and unzip it. The folder contains SKILL.md. Place the folder in one of these locations:
- Personal skills: ~/.claude/skills/ (unzip so the bgpt-paper-search folder sits inside it)
- Project skills: the .claude/skills/ folder inside a specific project
- Claude apps: upload the skill under Settings → Capabilities → Skills
Connect the BGPT server (required)
Installing the skill does not connect you to BGPT. You must configure the BGPT remote MCP server in your host app. The provider offers two endpoints: Streamable HTTP at https://bgpt.pro/mcp/stream and SSE at https://bgpt.pro/mcp/sse. Use your host's native remote-MCP settings when it supports them.
If your host uses an mcpServers JSON file and needs a local stdio bridge, the skill's documented example uses mcp-remote through npx. That requires Node.js and npm. Adapt the example to your host's actual settings format.
- Free access needs no key
- For paid usage, add an Authorization: Bearer <key> header in the host's remote-server headers
- With the mcp-remote bridge, you can supply the header from a local file using --header-file
- Do not put an API key inside a tool call; the current tools do not accept one
{
"mcpServers": {
"bgpt": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://bgpt.pro/mcp/stream", "--transport", "http-only"]
}
}
}Trigger the skill with natural requests
You do not need special syntax. Ask for what you need in plain language, and mention DOIs or the evidence you want. The agent should pick the skill when the request is about literature reviews, evidence synthesis, or experimental details beyond abstracts.
Keep search terms short and in English. Ask for filters such as recency or study type in the request itself and let the agent apply them, rather than writing years into the query.
- Find recent primary studies on CRISPR delivery in neurons and list their reported sample sizes
- Look up the paper with DOI 10.1016/j.ymeth.2015.10.014 and summarise its experiments and limitations
- Build a table of evidence on a question, with the DOI and source section for each claim
- Show me the query you ran and the DOIs you found so I can repeat the search
Read and verify the results
Results come back as a record with a list of results and a count, plus the query the service actually ran, since it may rewrite your terms. A DOI lookup returns whether the paper was found and the single record. Check for tool errors first, then check the record's extraction status and warnings.
Some records are older-format or contain no extracted claims, so a returned record does not guarantee claim-level detail. Each extracted claim is also bounded, so it is not the whole article. Before you use a finding, check the sample sizes, units, arms, statistics and uncertainty against the paper itself.
- Keep the DOI, title, publication date and publication name for every record
- Check extraction status and normalization warnings before synthesis
- Treat an empty result or a not-found DOI as inconclusive, not proof the paper does not exist
- Keep contradictory evidence and note the scope each paper demonstrates
Check pricing and limits
At the time of review the provider advertised 50 free results per network, then a charge per returned result. A successful DOI lookup counts as one result, and an unfound lookup costs nothing. Pricing can change, so check the provider page before running a large batch.
The provider did not publish a numeric request-rate limit. Do not assume unlimited throughput, and do not keep retrying a request once your allowance is used up.
- Run small test searches first to see how many results you use
- Use the num_results setting to limit how many results each search returns
What people use it for
- Literature reviews that need experimental details beyond abstracts
- Building a traceable evidence table for a student research question
- Checking the reported findings of a specific paper by DOI
- Comparing the scope and limitations reported across several papers on one topic
Frequently asked questions
Is the BGPT Paper Search skill free?
Free access needs no API key, and the provider advertised a free allowance of 50 results per network at the time of review. After that, results are charged per returned result. Check the provider page for current pricing before a large batch.
What do I need to use the bgpt-paper-search skill?
You need a configured remote BGPT MCP connection and internet access to bgpt.pro. If you use the mcp-remote stdio bridge, you also need Node.js and npm. A BGPT API key is only needed for paid usage.
How do I install the bgpt-paper-search skill?
Unzip the downloaded skill into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it in the Claude apps under Settings, Capabilities, Skills. Installing the skill does not connect you to BGPT, so you must also configure the MCP server.
Does the skill run scripts or send my data anywhere?
The skill does not run scripts, but it uses the network. Queries are sent to the external bgpt.pro service, so do not use it with confidential or unpublished data.
Can I use this skill for a systematic review?
Not as the only source. The skill states its records are not a complete search and not a validated risk-of-bias assessment, so use them to build a traceable evidence table rather than a formal systematic review.
Why did my search return a paper with no evidence?
Some returned records use an older format and contain an empty list of evidence units, so they do not guarantee claim-level detail. Check the extraction status and warnings, then read the original paper.
How does it compare with searching Google Scholar or PubMed?
The skill's focus is extracted evidence from full text, such as experiments, statistics, scope and limitations, rather than bibliographic metadata or abstracts. The source does not compare it with other databases, so test both on your own question.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you are a researcher, student or analyst who needs experimental details and reported statistics beyond abstracts, and you are comfortable verifying each finding against the paper. Pick a different tool for formal systematic reviews, risk-of-bias appraisal, or any work involving confidential or unpublished data, since queries go to an external service.
Download the skill