Skip to main content

Tool catalog

What each tool takes, returns, and costs.

MCP server URL

https://elevenflo.com/mcp
On this page

The live tools/list response is the exact input and output contract. Inputs reject unknown fields, and that rejection's recovery lists the arguments the tool takes. Errors return {code, message, retryable, recovery?, details?}, and public TextContent is the JSON copy of the same structuredContent payload. Every list-returning tool returns bounded rows; a limit or max_pages above a tool's maximum reads the maximum. ElevenFlo MCP exposes 16 tools: 12 bankruptcy research tools, three that manage the connected user's own monitors, and one that shows the connected account's plan and credits. This page is the complete list, and a tool absent here is not available. See Permissions and data access for what ElevenFlo MCP cannot do.

Cases are addressed by case_id, documents by UUID document_id. Use identifiers exactly as returned. ElevenFlo MCP accepts no aliases. When a document row carries has_primary_pdf: true, open the document at /api/search/open/<uuid>/ using that same UUID.

#The research loop

  1. find_cases to resolve a case_id.
  2. list_docket_entries, search_documents, or find_key_documents to select documents, and list_related_documents to follow one document to the motion, order, objections, replies or later amended version linked to it.
  3. search_document_pages to see which pages hold the keywords across up to 25 documents.
  4. read_document_pages for the text of those pages, or of the pages in one document that match a query.

#Cases and dockets

#find_cases

Resolve a case by identity. query matches case name, case number, jurisdiction and judge with typo tolerance; legal-form words such as Inc., LLC, Holdings and Group do not count toward a name match. It does not match counsel, docket text or document text (use search_documents for those). Screening is not this tool: filter or count a population with the cases structured dataset.

  • Inputs: query (required) and limit (1 to 50). Any other argument is rejected with invalid_input naming the argument, never an empty result.
  • Returns: case_id, identifying metadata, petition date, chapter, the petition asset and liability brackets with assets_basis_quality and liabilities_basis_quality (evidenced or legacy_unverified), document coverage and case-type tags. Best match first.

#list_docket_entries

List observed docket entries for one case, including full entry text, docket evidence, court citations, and links to held documents. Entries can exist without a PDF or searchable document text. For current case status, verify the relevant terminal notice rather than relying on a projected date.

  • Inputs: case_id; optional limit from 1 to 25, query, document_type_tags, inclusive date_filed_from/date_filed_to dates, up to 25 positive docket_entry_numbers or an inclusive docket_entry_number_from/docket_entry_number_to range (one bound alone is open-ended; the list and the range cannot be combined), and an opaque cursor from the preceding page. A cursor is valid only with the same filters. Dates describe filing dates, not independently verified entry or signature dates.
  • Text search: query matches an entry when every whitespace-separated term appears, in any order and ignoring case. When a date or docket entry number is known, start with those filters and omit query and type filters.
  • Filters: document_type_tags accepts only category keys and canonical tags; an unknown value returns a typed error listing the allowed values. Only entries with classified documents carry tags; entries known only from court RSS carry the court's short label and no tags, so tag and text filters can exclude them.
  • Returns: the case_id, entries, evidence and, when more entries follow, next_cursor; PDF/text flags describe held artifacts. Each entry carries its docket_entry_number, date_filed and docket_text, the court's own docket text in full. Every held document with a public UUID appears once, inside its entry, with its document_type_tags and a document_url that opens the document. A document repeats docket_text, date_filed, docket_entry_number or court_locator only where its own value differs from the entry's; a field it omits is the entry's. An empty first page carries coverage_note, which says whether the filters excluded recorded entries or none are recorded in that window, the case's coverage source, and whether the docket is known to be incomplete or its completeness is unknown.
  • Court locator: each entry and document can carry court_locator: the court citation (Bankr. D. Del., No. 22-11068, ECF No. 512; adversary and claim-register forms use Adv. No. and Claim No.), court, case_number, docket_entry_number or claim_number, case_docket_url, and an optional pacer link whose target is case, entry, or document. It says where the document sits on the court's docket, not where ElevenFlo's copy came from. A PACER link opens with the reader's own PACER login; PACER fees may apply. Document, summary, page and key-document results carry the same object.

#Monitors and case updates

#list_case_updates

Read what changed for the connected user's saved monitors, or for named cases, as a resumable feed. It reads; it does not schedule checks or send notifications.

  • Inputs: optional monitor_ids (from list_monitors) or one to 25 case_ids, never both; with neither it reads every saved monitor. Resolve a named case with find_cases first. Optional since is an ISO-8601 timestamp within the last 90 days for the first call; it defaults to seven days ago. limit is 1 to 50. Optional cursor continues or resumes; omit since with it.
  • Returns: changes in sequence order. An upsert carries the publication (headline, body, case, sources and citation ranges), its record status and the matching monitor_ids; a remove names a development that no longer matches, such as a corrected asset scope or a withdrawal. The first call pages a fixed snapshot; keep next_cursor and pass it later to get just the changes since. Every call returns next_cursor as a resume point; has_more is true when more changes are ready now. coverage states the scope, what unknown facts left out, and what the covered cases hold.
  • Limits: zero changes does not mean nothing was filed. After a monitor is edited, added or deleted the cursor returns monitor_changed: call again without cursor for a fresh snapshot. watched and triggers were removed: omit watched to read all your monitors, and set topics on a monitor with save_monitor instead of triggers. Monitor scopes need a signed-in user; market monitors need Pro.

#save_monitor

Save a monitor for the connected user, or edit one by monitor_id.

  • Inputs: filters: either case_id alone (a case watch) or market filters combined with AND: chapters, subchapter_v, court_states, statuses, event_kinds, asset_classes (a class includes its subclasses), min_units (requires asset_classes) and keyword (every word must appear). Optional name, and topics to narrow publication topics such as sale.
  • Returns: the monitor and a 28-day dry run: publications it matches and those left out because a fact a filter needs is unknown.
  • Limits: Free accounts hold five case watches; market monitors need Pro, up to 20. An edit keeps the id and makes earlier cursors return monitor_changed.

#list_monitors

List the connected user's monitors, oldest first, with their ids, filters, topics and filter_revision. limit is 1 to 100 (default 50); pass next_cursor back as cursor for the next page; it is absent on the last page. total counts every monitor.

#delete_monitor

Delete one of the connected user's monitors by monitor_id. Its feed and email updates stop. This cannot be undone.

These tools require every substantive query term to appear in the matching evidence. One common word cannot pull in unrelated results.

#search_documents

Search whole court documents or hearing transcripts across all cases or within one case. Every term must appear somewhere in the document, not necessarily in one passage. Quoted text is an exact phrase; uppercase OR, NOT and parentheses combine terms, for example "cash collateral" AND (objection OR NOT interim). A malformed query returns invalid_query naming the position to fix.

  • Inputs: query; optional content_type (filing, the default, or transcript), case_id, document_type_tags (not transcripts), inclusive date_filed_from/date_filed_to dates, and limit from 1 to 25.
  • Returns: document rows with document_id, document_url, case, docket_text (the court's docket text in full), docket_entry_number, date and a matching excerpt. Court documents come newest first and add type tags. Transcripts come in relevance order and add transcript_source_kind. An unknown document_type_tags value returns the allowed values.
  • The filing type venue_or_dismissal_motion covers motions to transfer venue, motions to dismiss a bankruptcy case and motions to dismiss an adversary complaint. Narrow by the relief sought, for example "motion to dismiss" AND 1112, and check the caption and proceeding. To include responsive filings and opinions, search the text across filing types.
  • A transcript is one of two kinds. certified_filing is a court reporter's transcript filed on the docket. Read it by page like other court documents. court_audio_asr is a machine transcription of court audio, not a court-reporter transcript; it can misstate names and figures and reads by time.

#Exact text

#search_document_pages

See which pages hold a few keywords across up to 25 selected documents or transcripts before reading full text.

  • Inputs: one to 25 document_ids and query.
  • Returns: hits, each a document_id, 1-based page and short snippet, for the pages that hold every keyword, in relevance order within a document. When a document has no such page, its closest pages come back with matched_terms. no_text_document_ids lists documents with no extracted text and missing_document_ids those not found or not public.
  • Limits: a document with no hit has no entry; that does not establish the topic is absent.

#read_document_pages

Read pages of one document or transcript.

  • Inputs: document_id and one of pages (numbers and ranges such as 3-5,9), query (a few keywords) or, for a court-audio transcript, time (such as 12:30); optional max_pages from 1 to 10 (default 3). With none of the three, the first max_pages pages are read.
  • Returns: each page's text and number, page_count and the document_url. For a numbered PDF page, page_url opens that page after sign-in. One call reads 10 pages. When pages names more, the first 10 in page order are read and unread_pages lists the rest that exist in the document, written so it can be passed as pages on the next call. With query, the pages holding every keyword come back in relevance order, all of them listed in matched_pages, with has_more_matches when more matched than were returned; if no page holds every keyword, the closest pages carry matched_terms. A court-audio transcript reads as numbered blocks of speaker turns with start_time and end_time; a court reporter's transcript reads by page, and time on one returns invalid_request.
  • When text is unavailable, source_not_searchable gives the original PDF link if ElevenFlo has a downloadable file. For a docket-only entry, it identifies the case for related-filing or hearing-transcript searches. Results from court audio should be cited as machine transcripts.
  • A document with text but no page map can be read by query. A page outside the document returns page_not_found; an unknown document returns a typed not-found error.

#Document relationships

#find_key_documents

Find key documents in a case, ranked by incoming citation count. Citation frequency is a starting point for research, not a judgment of legal importance.

  • Inputs: case_id, optional limit from 1 to 25.
  • Returns: document rows with citation counts, each document's docket_text and docket_entry_number, and a document_url that opens it.

List the documents one document's docket entry refers to and the documents whose docket entries refer to it, such as the motion an order rules on, the objections and replies to a motion, or the later amended version, redline notice or order whose docket entry cites an earlier filing.

  • Input: document_id.
  • Returns: two lists, references and referenced_by, holding up to 20 documents between them, newest filing first in each. Each list holds up to 10, and either list fills the room the other leaves. Each row has document_id, document_url, the court's docket_text, docket_entry_number, date_filed and relationship_type. truncated is true when more exist; the oldest in a full list are not listed.
  • Limits: a link exists where one docket entry cites another by number. It does not establish that a later filing amends or replaces the cited one. An empty list does not establish that nothing relates to the document.

#Structured data

Structured datasets carry typed, source-linked fields extracted from documents. Use them when the question is about typed fields, comparable records, or many cases at once rather than the text of one document. Responses return only public fields and operations, never internal extraction metadata.

Free accounts can list and describe datasets and run aggregates that do not name a case. Reading rows with search_dataset, and grouping by a field that names a case (case_id, id, case_name, case_number, lead_case_number, case_url), require ElevenFlo Pro.

#list_datasets

List every dataset, or count one case's records in each.

  • Inputs: optional case_id, a case identifier returned by find_cases.
  • Returns: columns once (dataset, one_row_per), then stages in the order a case runs. Each stage holds its families, and each family one array per dataset in the order of columns: the dataset name (the value the other dataset tools take) and what one row is. With case_id, each array ends with case_rows, the records the dataset holds for that case; search_dataset filtered on case_id reads them. A count of zero does not establish that an event did not occur.
  • The whole list is one credit. describe_dataset gives a dataset's title, freshness and coverage.

#describe_dataset

Call this before you query an unfamiliar dataset.

  • Inputs: dataset, a name returned by list_datasets; optional fields, up to 60 field names.
  • Returns: what one row is, as_of, complete_since, grade, cases_covered, caveats, one fields table and example calls. Each field row gives its name, type, unit, fill (the share of rows that hold a value), allowed values, the filter operators it accepts and its meaning. extra_filters lists filters that are not fields; sort and facets list the fields that sort and facet.
  • Every description is one credit. A wide dataset describes its main fields in fields and names the rest by type in more_fields. Pass any of those names as fields to get their rows; that response holds the dataset, its title, what one row is and the rows asked for, in the order asked.
  • case_fields lists the case facts every dataset except cases carries (chapter, court, judge, petition date, size ranges and the like) with each one's type and its allowed values when the list is short. They filter, select and group like the dataset's own fields; pass one as fields for its full row.
  • A filter has one name, the one describe_dataset lists; there are no aliases. Sale-process dates filter on latest_source_date_filed and fees-by-professional on latest_date_filed. Calendar-date ranges support eq, gt, lt, gte, lte, and between; numeric ranges support gte, lte, and between. exists tests whether a value is stated.

#search_dataset

Return typed rows from one dataset.

  • Inputs: dataset; optional typed where, select, sort, limit, cursor, and requested facets as listed by the dataset description.
  • Returns: columns once, then one array per record in that order, with total (null when not counted), facet counts when asked, next_cursor when more records match, as_of (null when unknown) and complete_since. Wide projections clamp page size to the select budget.
  • To find rows from a known filing, check which operators the dataset lists for document_id or docket_entry_number. Use eq for one filing or in for several. Pair docket entry numbers with case_id to identify the case. Datasets that require case_id also need it with a filing UUID. Both fields refer to the source filing identified on each row.
  • Stored spellings: to find how a name is stored, filter the field with contains and request it as a facet.
  • Limits: a cases screen on assets_range or liabilities_range cannot match a case whose petition reports no bracket.

#aggregate_dataset

Count and summarize records across one dataset.

  • Inputs: dataset and metrics; optional typed where, group_by, sort, and a limit of up to 1,000 groups. A date field groups by a time bucket and a number field by a fixed width. A metric's own where counts only the rows it matches, so two metrics in one call give a rate.
  • Returns: columns once, then one array per group: the group fields, each metric and, for a field metric, <metric>_n, the rows that held a value. <metric>_is_floor marks a sum that leaves out amounts the documents do not state, partial marks a time bucket that starts before complete_since, and totals gives each metric over every group.

#Available datasets

  • cases: one row per canonical bankruptcy case for screening by chapter, Subchapter V election, court and state, petition date, entity form and NAICS code from the lead petition, industry and case-type tags, petition creditor, asset and liability brackets with their evidence basis, stage, typed outcome and document coverage. Closed facts spell absence as unknown, missing or none, so a group-by on them counts every case once; tag group-bys count a case once per tag and overlap.
  • voluntary-petitions
  • top-unsecured-creditors
  • capital-instruments
  • capital-stated-totals
  • capital-parties
  • capital-links
  • dip-facilities
  • dip-orders
  • dip-cash-flow-budgets
  • dip-milestones
  • committees: one row per official committee with its current roster. notices lists every appointment notice with the roster that notice states, oldest first.
  • committee-members: one row per member's time on a committee. Filter is_current to true for today's roster. For the members on a date, filter joined_date with lte that date and on_committee_until with gt it. joined_date_basis and left_date_basis say whether a notice states the date or a filing date bounds it; count departures with end_transition equal to actual_departure.
  • schedule-summary-totals
  • schedule-assets: one disclosed asset item on a Schedule A/B; filter asset_category to real_property for real estate.
  • codebtors: one codebtor a debtor lists on Schedule H. Filter codebtor_is_case_debtor to true for codebtors whose name matches one debtor on the case's roster; false does not show a codebtor is outside the group.
  • codebtor-obligations: one schedule reference of one Schedule H row, joined to the Schedule D or E/F claim or Schedule G contract the same filing supports. case_id is required. link_status says how far the filing supports the link: line_match (the printed line number and the creditor name agree), name_match (a checked box and exactly one row for that creditor, or the one row whose description is the debt Schedule H notes after the creditor's name), line_match_name_differs, ambiguous, no_matching_row, schedule_not_read and no_reference carry no confirmed obligation. obligation_row_id is the id of the row in scheduled-claims or executory-contracts; one debt with several codebtors repeats on each, so count distinct obligation_row_id and do not sum obligation_amount_usd. Schedule H does not say whether a codebtor is a guarantor or a co-borrower; codebtor_debt_description_as_filed carries what a filing prints about the debt on Schedule H itself, when it prints anything. codebtor_is_case_debtor is true when the codebtor is itself a debtor on the case's roster. A debtor that files its schedules again has rows from each filing: filter is_latest_filing_for_debtor to true to read one filing per debtor, knowing a later filing can list only what it adds. A codebtor listed against several creditors or schedules still has one row per reference, so count distinct codebtor_row_id or codebtor names. candidate_lines gives the schedule line of each candidate row of an ambiguous reference. When the codebtor is another debtor of the case and the reference is to Schedule D or E/F, codebtor_own_row_status says what that debtor's own schedule lists for the creditor (one_row, several_rows, no_matching_row, schedule_not_read) and codebtor_own_row_id is its row in scheduled-claims: the same debt scheduled by a second debtor, never a second exposure. codebtor_own_row_basis says whether the row's description and lien read as the debt or it is only the one row naming the creditor, and codebtor_own_amount_agrees whether both debtors state one amount. scheduled_debtor_stated_role and codebtor_stated_role carry borrower or guarantor only where a debtor's own row prints it.
  • scheduled-claims: one claim listed on a Schedule D or E/F, with its bucket (secured, priority_unsecured, nonpriority_unsecured), stated amounts and disputed, contingent and unliquidated checkboxes. A secured claim's claim_nature_as_filed, collateral_description_as_filed and lien_description_as_filed are read apart: when one agent holds several facilities on the same collateral, the lien names the facility. An aggregate needs case_id or search.
  • sofa-disclosures: one answer to a Statement of Financial Affairs (Official Form 207) question, with its disclosure_type, party, amount, dates and the source text that states it. An aggregate needs case_id.
  • first-day-relief: one first-day motion or order with its relief type, requested and granted payment caps in their stated currency, and the rulings the orders state (granted_statuses).
  • bar-dates
  • claim-transfers
  • rule-2019-statements
  • rule-2019-memberships
  • rule-2019-positions
  • monthly-operating-reports
  • monthly-operating-report-periods
  • monthly-operating-report-case-periods
  • monthly-operating-report-coverage
  • keip-kerp
  • keip-kerp-terms: one row per payout trigger or payment condition of a program at one stage. Join program_id to a keip-kerp id. Compare the requested and authorized rows that share a term_label; order_effect says what the order did. Read award_amount with award_unit and award_scope. Select citations for the source pages. A program without rows has no terms available in this dataset; that does not show it has none.
  • asset-sales
  • case-assets
  • contract-treatment-events
  • executory-contracts
  • exclusivity-periods: one row per exclusive period a motion requests or an entered order grants, kept separately for filing a plan and soliciting acceptances. Order rows (source_kind order) carry the granted deadline, granted_days_from_petition, the linked motion, and the sequence in the case: prior_deadline_date, days_extended, extension_number and superseded_on_date. For the latest surviving grant on file on a historical date, filter order_filed_date lte and superseded_on_date gt that date. order_entered_date is available when the order’s court stamp explicitly gives an entry date. Motion rows carry the requested deadline and, when an entered order is linked, ruling_disposition and ruling_deadline_date. link_basis states how the motion and order were paired: docket_entry (the order's entry names the motion) or matching_deadline (the order is the first to grant exactly the requested deadline; an inference from the dates). A null sequence value or ruling means no link or an intermediate order is not on file; it does not show that the court did not act. days_granted_vs_requested on an order row is the granted deadline minus the linked motion's requested deadline; filter lte -1 for orders that gave less time than asked. An order a later order vacated has sequence_status vacated with vacated_on_date, and is never returned as the operative order. Count requests on source_kind motion: a related_paper row is an objection, statement or other paper about a motion, whose date can be another party's proposal; it names the motion it concerns and carries no ruling. objections_on_file on a motion row counts docket entries titled as an objection that cite the motion; 0 does not show the request was unopposed.
  • plans: each case's current Chapter 11 plan with lifecycle status, type, proponent, and filing, confirmation, and effective dates. A withheld date is null, never zero; an empty result never proves a plan event did not occur.
  • plan-classes: one row per class, with class-level recovery estimates and the published recoveries when available. recovery_mid_percent can be calculated from the class's bounds. Use the individual outcomes for elections.
  • plan-class-recoveries: one row per published recovery outcome. Filter by case_id. Use class_id to join the class terms and outcome and outcome_label to check which condition applies. Select citations to verify stated estimates and terms on the source page.
  • disclosure-solicitation-events
  • ballot-class-results
  • post-confirmation-reports
  • post-confirmation-report-quarters
  • post-confirmation-report-case-quarters
  • post-confirmation-report-coverage
  • post-confirmation-plan-comparison
  • conversions-dismissals-closures: court orders converting, dismissing or closing a case, newest-filed first. Rows default to rulings on the case itself; filter applies_to_tracked_case to no or unresolved for orders that end only other jointly administered cases or leave the scope open. affected_case_numbers lists the case numbers the order prints.
  • hearing-sessions: one row per current hearing session. A hearing_date sort is calendar order: one day's sessions follow the clock. For per-case totals group by case_id: the count is sessions, the sum of matter_count is agenda matters, and the minimum and maximum hearing_date are the first and last hearing.
  • adversary-proceedings: one row per adversary proceeding under a court number, with the parties and nature of suit the clerk announces on the bankruptcy docket and the dates of judgment, dismissal, closing and appeal entries read since. There is no status: a null date does not establish an open proceeding.
  • bankruptcy-appeals: one row per notice of appeal, with appellants, the docket entry numbers of the orders appealed, the court the appeal was first directed to, its case number there, the district judge where the district court's own docket is on file, and the ruling once the bankruptcy docket or that district docket records one. A null disposition does not establish a pending appeal; a panel's or court of appeals' docket is not collected.
  • counsel-appearances: one row per notice of appearance, pro hac vice motion or request for notice that names a client, a firm or an attorney: who represents whom, including creditors, lenders and ad hoc groups no retention order covers. A firm is recognized only when a retention, a fee application or the firm directory already names it. An appearance is not a court-approved retention, and a missing row does not show that a party is unrepresented.
  • case-people: one row per person, firm and case, for the people a case's documents name as working on it: what their firm does on the case, whom it acts for, the person's title, the latest document naming them and links to the person's and the firm's ElevenFlo pages. The firm is the one on that case, not current employment; count people with count_distinct on person_name.
  • professional-retentions: one row per professional engagement (counsel, financial advisers, bankers, claims agents and others) for a retaining party in a case, with its lifecycle status, the application and order it rests on, each objection, withdrawal and order in lifecycle_events, and stated rate tiers. Proposed is not approved; a status comes from an entered order.
  • professional-retention-rates: one row per stated rate tier. Compare like with like: filter band, period=hour and currency_code, and filter basis, because a requested rate is not an approved one.
  • professional-retention-fee-terms — compare monthly, sale and restructuring terms by firm, role, fee kind, source basis and currency. Select source_quote for exceptions and credits and source_pages for the filed evidence.
  • fee-applications
  • fee-decisions
  • fees-by-professional: cumulative requested and allowed fee economics by applicant, case, and currency; group by firm_name for requested fees by firm. Requested fees are not court-allowed fees. Affiliates appear separately without parent-firm consolidation, and roles may be unknown. The public requested_amount_rejected boolean indicates that requested economics exclude rejected documents; surviving requests in the same rollup still contribute. allowance_note states in plain words when allowed totals are blank, when a final basis also counts interim awards, or when combined or undated awards are left out or added unchecked. Monetary sorts require currency_code with eq; the default sort is latest filing date. sources identifies contributors separately for each monetary field, with availability, counts and truncation. Null provenance awaits reprojection. Order-date filters select applicants without rebasing their lifetime totals to the selected month.
  • fee-timekeepers: named historical billing rates by firm and case. Filter timekeeper_name with name tokens in any order, organization_name, timekeeper_title, or hourly_rate. Group partner rates by firm, filing year, and currency; aggregates default to monthly, final, and first-and-final applications. Rates are observed billing evidence, not current market rates.
  • fee-time-entries

This page is a static copy of the public catalog. Call list_datasets for the live list and describe_dataset for each dataset's fields, operations, and limits. Datasets outside the public catalog return unknown_dataset, and its recovery names the closest dataset names.

hearing-sessions uses schema hearing-sessions.v4. scope_kind distinguishes bankruptcy-case, single-adversary, joint-adversary and unresolved ownership. adversary_proceedings contains canonical court numbers, captions and available tracking URLs; scope_sources contains immutable public-document/hash references for qualified joint conferences. Unresolved scope does not assert case ownership. Select matter_sources for docket documents linked to the session's current matters. Each link names the matter and the filing's role, such as notice, adjournment or agenda. Pass its document_id to read_document_pages to check what the filing says about the setting.

ballot-class-results uses schema ballot-class-results.v3. It returns one row per plan class on each current ballot tabulation declaration, with the declaration's filing date, tabulation date, voting deadline and plan on every row. For one declaration's class counts, group by tabulation_id and reported_result and count rows. is_operative is true only for the latest non-supplemental declaration for the case and plan, or the latest supplemental one when the plan has no other. classes_complete is false when the declaration reports classes that are not rows. A declaration with no class row does not appear. These class rows are what the declaration reports; they are not canonical plan votes.

#Requests to ElevenFlo

Two tools send ElevenFlo staff something from the signed-in user, and only when the user asks: a request to add a case or missing documents, and a report about wrong data, an incorrect citation or a failed tool, or a suggestion. Neither costs credits. A prepaid tab has no user to answer, so neither is offered to one. Resending the same request or report returns the one already received, so a client can retry safely after a lost response.

#request_coverage

Ask ElevenFlo to add a case it does not list, or documents missing from a case it does.

  • Inputs: case_id from find_cases with missing_documents naming the filings, such as "Docket 412, the sale order". missing_documents is required when the case already has documents. For a case ElevenFlo does not list: case_name with court or case_number, and confirm_not_listed once the user has ruled out the possible matches a first attempt returned.
  • Returns: request_id, the reference to quote when following up; status (needs_review, then fulfilled or declined); created; documents_added; case_id once the case is identified; case_name; and a plain message. Staff email the user the outcome.
  • Members of one ElevenFlo account share one open request per case and can read its notes. A request naming another missing filing adds it to the open request.
  • If ElevenFlo lists a case matching the name or number, the tool returns case_listed or ambiguous_case with the matches and records nothing.
  • A request records interest. It does not promise that ElevenFlo can obtain the documents.

#submit_feedback

Report a problem with ElevenFlo data or tools, or make a suggestion.

  • Inputs: category (data_correction, incorrect_citation, tool_failure or suggestion), a description of up to 1,000 characters, and the references that locate the problem: case_id, document_id with page, dataset with row_id, tool_name, or request_id, the Request ID a failed tool returned. A data correction or incorrect citation names a case, document or dataset; a tool failure names a tool or Request ID.
  • Returns: feedback_id, the reference to quote when following up; status; created; and a plain message.
  • The report holds only these fields. ElevenFlo does not receive the conversation, prompts or credentials. A Request ID is accepted only from the user's own tool calls.

#Credits

MCP credits measure returned data. Each successful response costs the largest of 1 credit, returned records divided by 10 (rounded up), or canonical response bytes divided by 10,000 (rounded up). Protocol duplicates are counted only once. A 100-row response of 20 KB costs 10 credits; a 45 KB text response costs 5. Failed calls are not charged. Successful empty responses cost 1 credit. show_plan_and_credits, request_coverage and submit_feedback cost none.

Free includes 200 MCP credits per calendar month, with no daily limit. Pro includes 10,000 per calendar month and 2,000 per rolling 24 hours. Agents without an account buy prepaid tabs of 100 credits for USD 5.00; see Pay as an agent. All clients share the account balance and its 120-request-per-minute limit. Monthly balances reset at 00:00 UTC on the first of the month; Pro's rolling capacity recovers as usage ages past 24 hours. A result that exceeds remaining capacity, or the caller's elevenflo/max_credits request budget, is withheld and not charged. Every successful result reports its cost and the remaining balance in _meta. See current balances with show_plan_and_credits or on your account page.

#show_plan_and_credits

Show the connected account's plan and MCP credits.

  • Inputs: none.
  • Returns: plan (Free or Pro) and monthly_credits: the credits included, used and remaining this month and when they reset. Pro adds daily_credits for the rolling 24 hours. A Free account also receives upgrade: what ElevenFlo Pro includes, its price, and a link that opens the upgrade on elevenflo.com, where payment happens.
  • Costs no credits and works after a credit limit is reached. A client that supports MCP Apps shows the result as a panel in the conversation; any other client receives the same fields as text.
  • Limits: a prepaid tab holds a balance but no plan, so the tool is not offered to one.

Website document access has a separate rolling-day budget of 75 distinct documents on Pro. Free accounts keep their monthly download allowance instead. Reopening the same document and PDF range requests do not consume additional rolling-day slots.