--- name: papers-jobs description: "Step-by-step job recipes for the Paper Finder MCP server: Search papers; Search arXiv; Get a paper. Use when the user asks things like \"Find recent open-access papers on retrieval-augmented generation evaluation.\" or \"Search arXiv for papers on speculative decoding from this year.\"" license: MIT --- # Paper Finder jobs Each section below walks through one job with the tools of the `papers` MCP server. The same recipes are MCP prompts (prompts/list) on `https://papers.openkrill.app/mcp`. Connect the server at `https://papers.openkrill.app/mcp` in your MCP client settings (remote streamable HTTP, no key), then call its tools as normal tool calls. They are not command line programs: Bash, curl and scripts cannot reach them. ## Search papers Job: Search papers, with the Paper Finder MCP server. Use the `search_papers` tool for this job. Work on the user's request. If there is none yet, ask what they need. A request this job handles: "Find recent open-access papers on retrieval-augmented generation evaluation." Steps: 1. Call `search_papers` as a normal MCP tool call, with the request's values in the fields of its inputSchema. It needs `query`. 2. If a value the tool needs is missing and you cannot tell it from the request, ask the user one short question first. 3. This input is known to work; copy its format, not its values: `{"query":"systematic review intermittent fasting cardiovascular outcomes","from_year":2023}` 4. For the request "Find recent open-access papers on retrieval-augmented generation evaluation.", a correct answer looks like this: status ok and a merged list from Europe PMC and Crossref, each paper with title, authors, year, venue, DOI, an open-access link and the index it came from, with the source and as-of date. 5. Follow the server's own rules: - Use these tools to find research papers and their free full texts. search_papers asks Europe PMC and Crossref about a topic and merges the answers; search_arxiv searches arXiv preprints; get_paper returns one paper's abstract from a DOI, PubMed id or arXiv id. - Pass topic words, not a full question. - Use from_year for recent work and open_access_only when the user wants papers they can read for free. - Report each paper's source and tell the user that openAccessUrl null means the index does not confirm a free copy, not that none exists. - Read error.hint when a status is not ok and call again with a changed query instead of repeating the same call. - These are records, not endorsements: never present a paper's claims as established fact or give medical advice from an abstract. - Do not use these tools to check retractions or verify a reference list: say this plugin does not check them. 6. If the call fails or the answer looks wrong, tell the user, then report it with `submit_feedback` (kind bug, tool `search_papers`). ## Search arXiv Job: Search arXiv, with the Paper Finder MCP server. Use the `search_arxiv` tool for this job. Work on the user's request. If there is none yet, ask what they need. A request this job handles: "Search arXiv for papers on speculative decoding from this year." Steps: 1. Call `search_arxiv` as a normal MCP tool call, with the request's values in the fields of its inputSchema. It needs `query`. 2. If a value the tool needs is missing and you cannot tell it from the request, ask the user one short question first. 3. This input is known to work; copy its format, not its values: `{"query":"attention is all you need","limit":1}` 4. For the request "Search arXiv for papers on speculative decoding from this year.", a correct answer looks like this: status ok and preprints each with title, authors, year, the start of the abstract, the arXiv page and a PDF link. 5. Follow the server's own rules: - Use these tools to find research papers and their free full texts. search_papers asks Europe PMC and Crossref about a topic and merges the answers; search_arxiv searches arXiv preprints; get_paper returns one paper's abstract from a DOI, PubMed id or arXiv id. - Pass topic words, not a full question. - Use from_year for recent work and open_access_only when the user wants papers they can read for free. - Report each paper's source and tell the user that openAccessUrl null means the index does not confirm a free copy, not that none exists. - Read error.hint when a status is not ok and call again with a changed query instead of repeating the same call. - These are records, not endorsements: never present a paper's claims as established fact or give medical advice from an abstract. - Do not use these tools to check retractions or verify a reference list: say this plugin does not check them. 6. If the call fails or the answer looks wrong, tell the user, then report it with `submit_feedback` (kind bug, tool `search_arxiv`). ## Get a paper Job: Get a paper, with the Paper Finder MCP server. Use the `get_paper` tool for this job. Work on the user's request. If there is none yet, ask what they need. A request this job handles: "Get the abstract and open-access link for DOI 10.1038/s41586-021-03819-2." Steps: 1. Call `get_paper` as a normal MCP tool call, with the request's values in the fields of its inputSchema. 2. If a value the tool needs is missing and you cannot tell it from the request, ask the user one short question first. 3. This input is known to work; copy its format, not its values: `{"doi":"10.1038/s41586-021-03819-2"}` 4. For the request "Get the abstract and open-access link for DOI 10.1038/s41586-021-03819-2.", a correct answer looks like this: status ok with the paper's title, authors, year, venue, the start of the abstract, a free full-text link when the index confirms one, and the source and as-of date. 5. Follow the server's own rules: - Use these tools to find research papers and their free full texts. search_papers asks Europe PMC and Crossref about a topic and merges the answers; search_arxiv searches arXiv preprints; get_paper returns one paper's abstract from a DOI, PubMed id or arXiv id. - Pass topic words, not a full question. - Use from_year for recent work and open_access_only when the user wants papers they can read for free. - Report each paper's source and tell the user that openAccessUrl null means the index does not confirm a free copy, not that none exists. - Read error.hint when a status is not ok and call again with a changed query instead of repeating the same call. - These are records, not endorsements: never present a paper's claims as established fact or give medical advice from an abstract. - Do not use these tools to check retractions or verify a reference list: say this plugin does not check them. 6. If the call fails or the answer looks wrong, tell the user, then report it with `submit_feedback` (kind bug, tool `get_paper`).