MCP App Store
by KeenEthics
Hypha icon

Hypha MCP Server for ChatGPT

By Hypha

Overview

Hypha turns your organization's documents into structured, searchable data. Connect it to ask questions across everything you've uploaded — contracts, financials, reports — and get precise answers grounded in the actual facts pulled from those documents, with every answer traceable back to its source. Explore what Hypha has organized for your organization and pull exactly the details you need, from high-level summaries down to specific figures.

How to connect

  1. ChatGPT: Open the Hypha listing in the ChatGPT app directory and choose Connect. ChatGPT listing Hypha exposes 35 tools here.

Tools

accept_spread_mapping_drafts

ChatGPT
Accept reviewed reusable mapping drafts using the current statement draft fingerprint. Exactly one selector is required: candidateIds for up to 50 reviewed rows, cohortId for a complete reviewed import or local-definition cohort of any supported size, or acceptAll for the entire reviewed statement. Selected imported decisions replace their accepted baselines. Stale baselines and incomplete copies are rejected. Financial values and periods are outside this approval.

approve_all_financial_spread_documents

ChatGPT
Start asynchronous approval for up to 50 reviewed financial spread sheets from one source document. Each sheet requires its reviewed updatedAt version. Accepted and already_running outcomes describe workflow admission; read sheet approval status to confirm completion. Returns every sheet outcome with a short error summary; inspect approval readiness for conflict details.

approve_financial_spread_document

ChatGPT
Start asynchronous approval for one reviewed financial spread sheet. Requires its reviewed updatedAt version and computes the verification snapshot on the server. An accepted or already_running result means approval started or resumed, not that it completed.

create_financial_spread_adjustment

ChatGPT
Create one adjustment with up to 200 period values on a pending spread. Supply a new UUID adjustmentId and the reviewed expectedUpdatedAt. Reuse the same ID if the response is lost; GET it to reconcile a conflict. Read again after mutation before further edits. No approval is performed.

create_financial_spread_column

ChatGPT
Create one reviewed period mapping on a pending financial spread. Requires the spread updatedAt returned with the reviewed columns.

create_spread_mapping_definition

ChatGPT
Create a complete local draft mapping definition from the effective inherited statement using its reviewed definition fingerprint. The server copies all accounts, including explicit unmapped decisions. Browse the returned cohortId, edit in bounded batches, then accept that cohort with the current draft fingerprint. A refreshed copy has a new cohortId and leaves stale drafts unaccepted.

delete_financial_spread_adjustment

ChatGPT
Delete an adjustment and its cells from a pending spread using reviewed expectedUpdatedAt. Preserves carry-forward deletion audit history. Read again to reconcile a lost response; no automatic approval.

delete_financial_spread_column

ChatGPT
Delete one financial spread period mapping and its dependent raw values. Requires the full reviewed column snapshot and spread document ID for atomic stale-write and membership checks.

excel

ChatGPT
Read source Excel cells/ranges and query them with formulas, without modifying the workbook. Examples: =Sheet1!A1:D10, =MATCH("Revenue", A:A, 0), =VLOOKUP("Revenue", A:D, 4, 0). Use list_sheets for exact names and bounds. Direct reads preserve source display/raw values; calculations return calculated values. Outputs are bounded: narrow ranges on truncation. Cite [excel-document_id]{sheet_name-cell_ref}, or the sheet when cell_ref is absent.

get_entity_attributes

ChatGPT
Get one entity type's columns/attributes. Call after get_schema, once per entity type your question touches — several can run in parallel. Returns a lean outline of ALL the entity type's columns (column name, type, attribute group, and any lookup-table JOINs needed to resolve UUID foreign keys to display values) — everything you need to write SQL with run_sql. entityType must match a name exactly as shown in get_schema (e.g. "Loan").

get_financial_spread_adjustment

ChatGPT
Read an adjustment, its source anchor, note, target and every period value through nextCursor. Restart paging if spreadDocumentUpdatedAt changes. mcpTruncation.truncatedFields identifies shortened display fields; do not write those shortened strings back.

get_financial_spread_approval_readiness

ChatGPT
Return approval blockers, warnings and workbook-backed pre-approval footer tie-outs. Calculation failures block readiness; mismatches and deliberately unmapped footers are warnings, so ready does not certify mapping correctness. Default comparisonView=cells returns source/computed values per period; comparisonView=rows retains every mapped footer with total source periods, matched/mismatched/failed/uncompared counts and targetColumnName. Unsupported source roles count as uncompared, not failures. A supported-period tie-out requires matched=total-uncompared; inspect uncompared evidence separately under your terminal-metric policy. Rows with only uncompared periods remain visible with mismatchesOnly=true. comparisonView=coverage returns up to 10 footerCoverage entries per page, including unmapped/no-period/failed footers and the count of periods actually compared. Blank source versus computed zero matches, while missing computed results remain failures. Set mismatchesOnly=true to filter before pagination (includes failed and unverified checks). Returns up to 25 cells or row summaries; pass comparisonOffset=aggregationCheck.nextOffset with the same view/filter for more. resultCount covers the selected view/filter; comparisonCount and mismatchCount always cover the whole sheet. absoluteTolerance defaults to 0 (strict); use 2 for integer-rounded currency sheets. percentageTolerance defaults to 0 (raw decimal units). Compare fingerprints across pages and restart if changed.

get_schema

ChatGPT
Get the organization's data model as a compact tree: entity types, their tables, and how they relate (1:N / N:M join keys), plus the document-type catalog. Call this FIRST, once. It is the map — it does NOT list columns. For an entity type's columns/attributes, call get_entity_attributes({ entityType }). Use the relationships and join keys shown here to plan JOINs before writing SQL with run_sql.

import_spread_mapping

ChatGPT
Start mapping-only extraction from an authorized source XLSX document into the selected mapping scope and configured statements. Returns a progressId; inspect list_spread_mapping_imports, review draft candidates, and explicitly accept. Does not publish financial spreads, values or periods. An active extraction is reported as a conflict; inspect its progress before retrying.

list_entities

ChatGPT
Discover current readable entities beneath an explicitly selected root across multiple entity types. Returns compact identities and current type names/kinds; requires loan:read and checks each entity’s native authorization root. Continue with nextCursor until null; restart on a changed hierarchy/visibility snapshot. entityTypeId filters output while traversing intermediate types. excludeEntityKinds and excludeEntityTypeIds prune entire branches. No attributes, hidden ancestors, or private authorization roots are returned. Mapping eligibility is separately checked by mapping definitions.

list_financial_line_items

ChatGPT
List each persisted line item once for one spread document, with mapping state and a shared preview of the latest three reported columns by default (up to five). Legacy spreads with no reported-role columns fall back to legacy-unclassified columns. Source values come from the workbook cells used by Spread View; formulas are opt-in. Use get_entity_attributes with purpose="modify" to resolve writable mapping target IDs.

list_financial_spread_adjustments

ChatGPT
List adjustment metadata and the reviewed sheet version. Page through nextCursor; restart paging if spreadDocumentUpdatedAt changes. Use get_financial_spread_adjustment for values. mcpTruncation.truncatedFields identifies shortened display fields; do not write those shortened strings back.

list_financial_spread_columns

ChatGPT
List every persisted spread column, including source headers, period naming, scenario evidence, stable IDs, and the sheet status and version. Review pending_approval or approved sheets; processing results are provisional. When paging, restart if columnsFingerprint or spreadDocumentUpdatedAt changes between pages.

list_financial_spread_documents

ChatGPT
Discover financial spread sheets and their stable spreadDocumentId values. Filter directly for sheets needing approval and/or mapping. Each result includes the separate source documentId, mapping counts, parent entities, and the exact UI path.

list_financial_spread_parent_candidates

ChatGPT
List the entities eligible to become the parent entity for a column in this spread document. Pages request 50 candidates to limit repeated graph traversal. Candidate IDs are validated again when a column is created or updated.

list_sheets

ChatGPT
List sheets and data bounds in an uploaded Excel workbook or PDF table workbook. Discover document paths with workspace_browse. Sheet names are exact inputs for excel; follow nextOffset for additional sheets.

list_spread_mapping_definitions

ChatGPT
Use list_entities with the selected root to find readable owner candidates, then discover statement mapping definitions, inheritance status, approved definition fingerprints and local draft fingerprints. Use the definition fingerprint with create_spread_mapping_definition to create a complete local draft definition and the draft fingerprint to explicitly accept reviewed candidates.

list_spread_mapping_documents

ChatGPT
Discover current eligible XLSX source documents linked within this mapping context. These IDs can start mapping-only imports.

list_spread_mapping_imports

ChatGPT
Inspect mapping-only extraction runs and durable receipt-backed completion. Completed imports produce draft candidates. Review and accept those separately; extraction completion never approves mappings or financial values.

list_spread_mapping_rules

ChatGPT
Inspect exact reusable mapping accounts, decisions, provenance and reviewed base IDs for one statement. Approved inherited rules and local drafts are separate. Read all pages before accepting an entire statement. Preserve exact source labels, paths and codes for edits.

list_spread_mapping_targets

ChatGPT
List live targets for one configured financial statement. Data rows use spread attributes; aggregation rows use computed or totaling attributes. Use target IDs from this bounded catalog.

map_financial_line_items

ChatGPT
Apply validated financial line-item mappings to one spread document. Every row carries the updatedAt value returned by list_financial_line_items for stale-write protection. Returns independent per-row outcomes and recalculation queue status.

reopen_financial_spread_document

ChatGPT
Reopen one approved financial spread sheet for mapping edits. Requires the reviewed updatedAt version. Preserves approved_at and published entity and attribute values while atomically returning raw line-item values to the pending state.

run_sql

ChatGPT
Execute a read-only SQL SELECT query against the organization's structured data. MANDATORY: Call get_schema({}) FIRST to learn tables + relationships, then get_entity_attributes({ entityType }) for a table's columns. Never guess table or column names. Rules: - PostgreSQL SELECT only (no INSERT/UPDATE/DELETE). - No schema prefixes — search_path is pre-configured (exception: reference tables use the schema-qualified names shown in get_schema). - Auto-LIMIT 500 is added if you omit it (aggregations don't need a LIMIT). - Use full table names, not aliases. - JOIN parent → child on the keys shown in get_schema: 1:N via the child's FK column; N:M via the junction table. For UUID foreign keys, LEFT JOIN the lookup table to show display values, not raw UUIDs. - Escape single quotes by doubling them (O''Brien). Returns rows as JSON objects for code and a markdown table for the conversation. Full docs: https://app.hypha.co/docs/mcp

save_spread_mapping_drafts

ChatGPT
Create or edit up to 50 reusable mapping draft decisions. Each row requires its reviewed expectedBaseMappingId, or null for a new local account. replacesCandidateId identifies an imported draft to replace. explicitUnmapped=true records intentional unmapping. The first child customization requires a complete local draft definition. Drafts have no runtime effect.

update_financial_spread_adjustment

ChatGPT
Update adjustment metadata and/or period values atomically on a pending spread using reviewed expectedUpdatedAt. Omitted fields and periods stay unchanged. valueRaw null clears a cell; zero sets zero. Invalid numbers or duplicate periods reject the whole request. Read again after mutation.

update_financial_spread_column

ChatGPT
Relink or correct one financial spread period mapping. Requires the full reviewed column snapshot and spread document ID for atomic stale-write and membership checks.

workspace_browse

ChatGPT
Discover uploaded source documents by name, type, entity or document IDs. Paginated metadata, not extracted values or corpus counts. Use returned paths with workspace_grep/workspace_read, or list_sheets/excel for workbooks. Results are bounded; follow nextOffset and truncation flags.

workspace_grep

ChatGPT
Search literal text in uploaded OCR documents (case-insensitive by default). Each call scans at most 20 documents and 2 million characters per document. Returns line/page locators and source URLs. Cite sources as Markdown links using the returned url; path is a tool locator, not a browser URL. Excel files are listed separately for list_sheets/excel. Partial results, skipped documents or nextOffset mean the search is incomplete; narrow document_ids and use workspace_read to continue within a document.

workspace_read

ChatGPT
Read uploaded OCR source text with L-prefixed line numbers, page locator and document citation. Use mode lines with start_line/end_line for targeted reads. Full reads are bounded to 5000 UTF-8 bytes per result; Continue truncated reads with start_line=next_line and start_character=next_character; offsets count UTF-16 units and may remain within the same line. Excel documents use list_sheets/excel. Cite sources as Markdown links using the returned url with the document name and page; citation is the machine line locator. The url opens the first cited page when known.

Capabilities

Writes

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/hypha"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/hypha)