regulations_browse
ChatGPTBrowse the table of contents of 21 CFR. With no part it returns every part held and how many sections each has; with a part it returns that part's sections and their headings in document order. Headings only, never regulation text, so this is the cheap way to find which section a topic lives in before reading it with regulations_getSection. Reach for it when a question is about a whole rather than a phrase — how many sections a part has, every section in a subpart, the complete list of something — because regulations_search returns a ranked sample and a sample can never tell you what a complete set contains, so a count taken from it is confidently wrong; browse returns the whole contents in document order, so you count from the text and then read the section that matters with regulations_getSection. Reach for it too when regulations_findText has found nothing and you need to see what is there. Every result carries a coverage field naming the parts and the annual edition the corpus actually holds, and every row carries an eCFR url built against that edition. A part the corpus does not hold comes back as not found, naming the parts that do exist. There is no paging here: a part returns all of its sections in one call.
regulations_findText
ChatGPTFind passages of 21 CFR whose text or heading contain a phrase, each returned with its citation and an eCFR url so you can quote the regulation exactly. This is a literal substring match and nothing more: it is not ranked, not semantic, and will not find synonyms or reworded phrasing, so a run of this is not a search of the regulations and an empty result is not evidence a rule does not exist. Prefer a distinctive phrase from the regulation itself over a natural-language question, try a shorter phrase before concluding anything, and use regulations_browse to see what is there when a phrase finds nothing. Matching is case-insensitive. Optionally narrow to one part. Results come back in document order, not by relevance, and every result carries a coverage field naming the parts and the annual edition the corpus actually holds, so scope never has to be assumed. Passages are long, so ask for few and page with nextCursor; isPartial says when a further page exists.
regulations_getCitation
ChatGPTRead one passage of 21 CFR by its exact id, as returned in the id field by regulations_findText or regulations_getSection. The id looks like "21 CFR 101.9(c)(1)" and sometimes carries an occurrence suffix such as "#2", which is part of the id and must be kept. This does not search: an id you have assembled yourself rather than copied will usually come back not found. Use regulations_getSection when you have a section number rather than a passage id. Returns the passage with an eCFR url pointing at the edition the corpus holds, plus a coverage field naming the parts and edition it holds.
regulations_getSection
ChatGPTRead a section of 21 CFR in document order, so you can follow a rule through its subsections rather than quoting a fragment. Read it whole because exceptions live next to the rules they modify: a paragraph read on its own can say the opposite of what it means in place. So after a regulations_search hit that starts mid-thought, pull its section here to see the subsection it sits in. Give a paragraph rather than a section to narrow the read to that paragraph and everything nested under it: "101.9(c)(1)" returns (c)(1) and its children, not the rest of (c). Use this once regulations_search, regulations_findText or regulations_browse has told you which section matters, or when you already know it. Each passage comes back with its citation and an eCFR url, and the result carries a coverage field naming the parts and annual edition the corpus actually holds. Long sections page: 101.9 alone runs to over two hundred passages, so check isPartial and nextCursor before treating a page as the whole section. A citation that is not a section reference comes back as invalid input; a section or paragraph the corpus does not hold comes back as not found, saying which of the two it was.
regulations_search
ChatGPTSearch 21 CFR by the words of a question and by its meaning at once, and return the passages that best answer it, each with its section citation, an eCFR url, matchType and similarity. This is the tool to reach for first on any question about what the regulations require, permit or define. It runs two searches and combines them, which is why it finds a rule whose wording differs from yours and also finds a section named by its exact term of art; regulations_findText is the narrower literal alternative, worth reaching for only to confirm an exact phrase appears. Ask in regulatory language where you can, because it retrieves better: "reference amount customarily consumed for breakfast cereal" beats "how much cereal is a serving". matchType says which search found a passage — words, meaning, or both — and both is the strongest signal here, because the terms are present and the meaning is near. similarity is how close the meaning search sat, and is null for a passage only the word search matched; it measures resemblance, not correctness, so a high one is not evidence the passage answers the question and the citation still has to be read. This answers a question about a phrase or a value; a question about a whole — how many categories exist, every claim type, all the requirements in a subpart — is not one a ranked sample can settle, so browse the contents with regulations_browse and read the section whole instead. Results are passages, not whole sections, so follow up with regulations_getSection when a rule needs its surrounding subsections. The coverage field names the parts and annual edition actually held, so scope never has to be assumed, and a rule outside those parts will not be found however it is phrased. An empty list is not an error and is not proof the rule does not exist.
guidance_getDocument
Claudeguidance_search
Clauderegulations_browse
ClaudeBrowse the table of contents of 21 CFR. With no part it returns every part held and how many sections each has; with a part it returns that part's sections and their headings in document order. Headings only, never regulation text, so this is the cheap way to find which section a topic lives in before reading it with regulations_getSection. Reach for it when a question is about a whole rather than a phrase — how many sections a part has, every section in a subpart, the complete list of something — because regulations_search returns a ranked sample and a sample can never tell you what a complete set contains, so a count taken from it is confidently wrong; browse returns the whole contents in document order, so you count from the text and then read the section that matters with regulations_getSection. Reach for it too when regulations_findText has found nothing and you need to see what is there. Every result carries a coverage field naming the parts and the annual edition the corpus actually holds, and every row carries an eCFR url built against that edition. A part the corpus does not hold comes back as not found, naming the parts that do exist. There is no paging here: a part returns all of its sections in one call.
regulations_findText
ClaudeFind passages of 21 CFR whose text or heading contain a phrase, each returned with its citation and an eCFR url so you can quote the regulation exactly. This is a literal substring match and nothing more: it is not ranked, not semantic, and will not find synonyms or reworded phrasing, so a run of this is not a search of the regulations and an empty result is not evidence a rule does not exist. Prefer a distinctive phrase from the regulation itself over a natural-language question, try a shorter phrase before concluding anything, and use regulations_browse to see what is there when a phrase finds nothing. Matching is case-insensitive. Optionally narrow to one part. Results come back in document order, not by relevance, and every result carries a coverage field naming the parts and the annual edition the corpus actually holds, so scope never has to be assumed. Passages are long, so ask for few and page with nextCursor; isPartial says when a further page exists.
regulations_getCitation
ClaudeRead one passage of 21 CFR by its exact id, as returned in the id field by regulations_findText or regulations_getSection. The id looks like "21 CFR 101.9(c)(1)" and sometimes carries an occurrence suffix such as "#2", which is part of the id and must be kept. This does not search: an id you have assembled yourself rather than copied will usually come back not found. Use regulations_getSection when you have a section number rather than a passage id. Returns the passage with an eCFR url pointing at the edition the corpus holds, plus a coverage field naming the parts and edition it holds.
regulations_getSection
ClaudeRead a section of 21 CFR in document order, so you can follow a rule through its subsections rather than quoting a fragment. Read it whole because exceptions live next to the rules they modify: a paragraph read on its own can say the opposite of what it means in place. So after a regulations_search hit that starts mid-thought, pull its section here to see the subsection it sits in. Give a paragraph rather than a section to narrow the read to that paragraph and everything nested under it: "101.9(c)(1)" returns (c)(1) and its children, not the rest of (c). Use this once regulations_search, regulations_findText or regulations_browse has told you which section matters, or when you already know it. Each passage comes back with its citation and an eCFR url, and the result carries a coverage field naming the parts and annual edition the corpus actually holds. Long sections page: 101.9 alone runs to over two hundred passages, so check isPartial and nextCursor before treating a page as the whole section. A citation that is not a section reference comes back as invalid input; a section or paragraph the corpus does not hold comes back as not found, saying which of the two it was.
regulations_search
ClaudeSearch 21 CFR by the words of a question and by its meaning at once, and return the passages that best answer it, each with its section citation, an eCFR url, matchType and similarity. This is the tool to reach for first on any question about what the regulations require, permit or define. It runs two searches and combines them, which is why it finds a rule whose wording differs from yours and also finds a section named by its exact term of art; regulations_findText is the narrower literal alternative, worth reaching for only to confirm an exact phrase appears. Ask in regulatory language where you can, because it retrieves better: "reference amount customarily consumed for breakfast cereal" beats "how much cereal is a serving". matchType says which search found a passage — words, meaning, or both — and both is the strongest signal here, because the terms are present and the meaning is near. similarity is how close the meaning search sat, and is null for a passage only the word search matched; it measures resemblance, not correctness, so a high one is not evidence the passage answers the question and the citation still has to be read. This answers a question about a phrase or a value; a question about a whole — how many categories exist, every claim type, all the requirements in a subpart — is not one a ranked sample can settle, so browse the contents with regulations_browse and read the section whole instead. Results are passages, not whole sections, so follow up with regulations_getSection when a rule needs its surrounding subsections. The coverage field names the parts and annual edition actually held, so scope never has to be assumed, and a rule outside those parts will not be found however it is phrased. An empty list is not an error and is not proof the rule does not exist.