Maven Bio MCP Server for Claude and ChatGPT
By Maven Bio
Overview
Maven Bio gives Claude real-time structured life sciences data. Drug products, companies, clinical trials, indications, deals, financings, mechanisms, and targets, linked to each other and to the source documents behind them. Ask in plain language. Claude resolves the entity, pulls canonical records, then reads the primary source. "AZD6738" and "ceralasertib" land on the same drug. A landscape query returns the programs in a given indication. A claim about an approval traces back to the FDA letter itself. The Maven Bio connector enables: - Resolving messy names, aliases, and internal codes to canonical entities - Proactively monitoring opportunities: FDA actions, trial readouts, and press releases scoped to the entities a monitor tracks, each signal carrying its source - Discovery by criteria: "Phase 3 KRAS G12C inhibitors," "public oncology companies with late-stage pipelines" - Competitive landscaping by indication, grouped by phase, company, or mechanism - Evidence-based analysis: Search and read FDA filings, SEC filings, clinical papers, press releases, and conference abstracts as full text, structured sections, or extracted citations - Deal benchmarking: Public-company financials, financing rounds, and BD deals filterable by indication - Counts and aggregations across the entity graph Maven Bio MCP server authenticates every call, enforces read-only access and response limits. Users sign in with OAuth 2.1 against their existing Maven Bio identity.
How to connect
- Claude: Open the Maven Bio listing in the Claude connectors directory and choose Connect. Claude listing Maven Bio exposes 16 tools here.
- ChatGPT: Open the Maven Bio listing in the ChatGPT app directory and choose Connect. ChatGPT listing Maven Bio exposes 16 tools here.
Tools
fetch_related
ChatGPTunresolved array. In a mixed batch, inputs whose type cannot traverse to related_type are skipped and reported in a top-level errors array while the valid inputs are still processed; the call only errors outright when no input can traverse. Supports modality as an input type for modality-based queries (e.g., ADC -> companies). This is an input-only type and cannot be used as related_type. For DISCOVERING entities by criteria or attributes, use search_entities instead.get_artifact
ChatGPTget_deals
ChatGPTcompany matches any party to the deal (in any role); pair it with role to scope that company to a specific side of the deal (e.g. company=['Pfizer'], role=['Licensor'] returns deals where Pfizer is out-licensing). company accepts a list (OR'd) of either names or co_... IDs (mutually exclusive with company_id -- see per-field docs). deal_type accepts canonical types ('Licensing', 'R&D collaboration', 'Manufacturing collaboration', 'Commercialization agreement', 'Joint venture', 'M&A - company', 'M&A - product/asset', 'Spinoff', 'Platform/technology access', 'Service agreement', 'Other') plus common aliases ('M&A' -> 'M&A - company', 'JV' -> 'Joint venture', 'licensing deal' -> 'Licensing'); matching is case- and whitespace-insensitive. role accepts canonical roles ('Licensor', 'Licensee', 'Collaborator', 'Acquirer', 'Acquiree', 'Investor', 'Investee', 'Manufacturer', 'Distributor', 'Sponsor', 'Partner', 'Seller', 'Service Provider', 'Customer') plus aliases ('buyer' -> 'Acquirer', 'out-licensor' -> 'Licensor', 'in-licensee' -> 'Licensee'). deal_subtype accepts canonical subtypes ('Option agreement', 'Exclusive license', 'Non-exclusive license', 'Profit/cost sharing', 'Royalty-based', 'Distribution', 'Co-promotion', 'Supply', 'Other'); matching is case-insensitive. deal_status filters by deal lifecycle ('active', 'proposed', 'terminated', 'withdrawn', 'completed'). indication accepts indication or broad-area name(s) (e.g. 'NSCLC', 'oncology', 'immunology'). Specific diseases resolve onto the hierarchy-expanded indication filter (a parent matches itself and its children); broad-area terms resolve onto the curated disease_area facet where they are not hierarchy nodes. For a broad commercial area, prefer the disease_area parameter (25 areas such as 'Immunology & Inflammation', 'Oncology, Hematologic', 'Cardiometabolic', with aliases like 'immunology' and 'heme onc'): its boundaries follow how BD teams buy (immunology excludes the hematologic-malignancy branch that raw subtree expansion drags in), and it ANDs with indication for queries like 'immunology deals in ulcerative colitis'. Resolved terms are reported under resolved_filters.indication / resolved_filters.disease_area. Note ~76% of recent licensing deals carry an indication tag (M&A / distribution deals often have none), so untagged deals may be missed. Partial name/type/role resolution surfaces resolved_filters (companies: input -> {canonical name, co_... ID}; deal_type / role: input -> canonical) and unresolved_filters (per company role: [{input, candidates: [{name, score}, ...]}], up to 5 near-matches). If every company name is unresolved the call errors with entity_not_found; if every deal_type or every role is unresolved it errors with invalid_parameter and the canonical list. Each deal includes its parties as {id, name, role} pairs plus linked drugs, indications, trials, and supporting document_ids for chaining into read_document. value_status disambiguates the USD value fields: "confirmed" means Maven has a USD headline amount (in total_deal_value_usd). "unconfirmed" means Maven's data does not have the amount (USD fields are null); the deal may still have a publicly disclosed value Maven has not captured -- verify via the linked document_ids if the amount is critical. Defaults to event_date desc. For 'biggest first', pass sort_by='total_deal_value_usd' (NULL values ordered last). detail_level controls row weight: 'compact' (default) omits the multi-paragraph description and synergies` narrative fie…get_financials
ChatGPTmetrics is a projection: the response carries only the fields you request, plus fiscal_date/period/currency. Omit it for everything. Each metric maps to specific fields: - revenue -> revenue_usd - ebitda -> ebitda_usd - profitability -> cost of revenue, gross profit, operating income, net income, EPS - opex -> R&D and SG&A expense - balance_sheet -> assets, liabilities, equity, debt, cash - cash_flow -> operating/investing/financing cash flow (balance sheets are a separate metric) - market_cap, analyst_ratings Statement families are filed on different schedules, so latest periods can differ across income statements, balance sheets, and cash flows. Each company carries a statement_freshness block reporting the latest period per family. Use this tool when comparing financials across multiple companies, analyzing time-series trends, or building financial models. The default compact response returns latest market cap instead of long daily market-cap history; set detail_level='full' when you need market_cap_history. Full history is daily only for one company over one year; longer or multi-company calls are downsampled.get_financings
ChatGPTrecipient is the funded company; investor is a participating investor. Each role accepts a list (OR'd within a role) of either names or co_... IDs (mutually exclusive -- see per-field docs). financing_type accepts canonical types ('Pre-Seed', 'Seed', 'Series A', 'Series B', 'Series C', 'Series D+', 'IPO', 'Post-IPO Equity', 'Debt Financing', 'Grant', 'Strategic Investment', 'Acquisition', 'Other Financing') plus common aliases ('Series D' -> 'Series D+', 'M&A' -> 'Acquisition', 'post ipo equity' -> 'Post-IPO Equity'); matching is case- and whitespace-insensitive. Partial name/type resolution surfaces resolved_filters (companies: input -> {canonical name, co_... ID}; financing_type: input -> canonical) and unresolved_filters (per role: [{input, candidates: [{name, score}, ...]}], up to 5 near-matches; empty for typos with no substring overlap -- fall back to match_entity / search_entities). If every name in a role is unresolved the call errors with entity_not_found and the same candidates under alternatives; if every financing_type is unresolved it errors with invalid_parameter and the canonical list. Each round includes recipient/investor companies as {id, name} pairs (investor list may include display_only stubs -- resolvable via research_entity with limited data) and supporting document_ids for chaining into read_document. value_status disambiguates total_value_usd: "confirmed" means Maven has a USD amount for the round (in total_value_usd). "unconfirmed" means Maven's data does not have the amount (total_value_usd is null); the round may still have a publicly disclosed value Maven has not captured -- verify via the linked document_ids` or external sources if the amount is critical to your analysis. Defaults to financing_date desc. For 'biggest first', pass sort_by='total_value' (NULL values ordered last).get_monitors
ChatGPTget_recent_events
ChatGPTcontext when needed. 2. Monitor-scoped: pass monitor_id to see events across every entity the monitor tracks, dashboard rows and alert-only items alike. 3. Broad: omit both for Maven's full event feed. Use event_types to filter (e.g. ["fda", "trial"] for regulatory and clinical only). Results are reverse-chronological (newest first). Financing rounds are surfaced via get_financings; deal announcements via get_deals. To page through results, pass the response's next_cursor back as cursor. Do not build a cursor from dates: after_date and before_date are time-window filters, not pagination controls, and each call keeps whatever window you set. For a bounded window, combine after_date (oldest included) and before_date (newest included). Both are inclusive YYYY-MM-DD bounds on the event date.list_artifacts
ChatGPTmatch_entity
ChatGPTname and the required entity_type. A program is the exception: it is a drug in one indication and line, so it has no name of its own. Pass a prog_ id in name, or the drug name with the indication in context (e.g. name="Keytruda", context="NSCLC", entity_type="program"). When that pair covers several lines, the error lists each one with its id so you can pick. If you have extra identifying context such as sponsor, company, or a disambiguating note, pass that in context rather than packing it into name. Use search_entities when you are discovering entities by criteria or natural-language filters. Use research_entity after you already know the canonical entity you want to profile. Financings have no canonical short name; they are not resolvable here. Use search_entities(entity_type="financing", query=...) for criteria-based discovery or get_financings(...) for precise filters by recipient, investor, type, value, or date.read_document
ChatGPTresearch_bio_evidence
ChatGPTaspects to control what is returned (omit for sensible defaults per entity_type). Valid aspects depend on entity_type: - target: associations, tractability, constraint, known_drugs, repurposing (default: associations, tractability) - disease: associations, gwas, repurposing (default: associations) known_drugs is a TARGET-side aspect: it lists the drugs in development AGAINST the target you named. It is not a drug lookup. Takes a target or a disease, not a drug. For a drug's mechanism, targets, or indications use research_entity; for the drugs hitting a target use entity_type='target' with aspects=['known_drugs']. For a specific target-disease pair, pass entity_type='target' with the disease in context. Repurposing hypotheses are opt-in via aspects=['repurposing'] and are always speculative. Each result carries an evidence_rating (Direct/Indirect/Minimal). Database-curated associations are Direct; inferred scores are Indirect; a fuzzy disease-ontology match is downgraded one level. Not for clinical trial data (use research_landscape) or financials (use get_financials).research_entity
ChatGPTname -- Maven resolves variants internally (e.g. "pembro", "Keytruda", "pembrolizumab" all work). Do NOT pass IDs here; pass names. The one exception is program, which has no name: pass a prog_ id in name, or the drug name with the indication in context. See match_entity. If you have extra identifying context such as a sponsor, company, or disambiguating note, pass that in context rather than packing it into name (e.g. name="Tenecteplase", context="Genentech"). Use aspects to request specific data sections. Omit for the default aspects. Valid aspects depend on entity_type: - drug: trials, events, documents, indication_phases - company: trials, events, documents, financials [+ opt-in: financings, investments, deals] - trial: events, documents - mechanism: trials, events, documents, indication_phases - target: trials, events, documents, indication_phases - program: development_success (opt-in; the program's only aspect) Opt-in program aspect: development_success returns a published historical rate at which programs entering this program's phase in its disease area went on to approval. It is a cohort base rate, not a forecast for this asset and not a probability conditional on the current phase already being underway. Where the indication spans several of the source's disease areas the rate combines those cells and says which. When the program cannot be matched to a cell the block states why instead of returning a number. Opt-in company aspects: financings returns funding rounds where the company is the recipient; investments returns rounds where the company is an investor (useful for VCs, corporate venture arms, and investor companies); deals returns licensing, M&A, and collaboration deals where the company is any party, each row carrying a per-party role. All require explicit opt-in via the aspects argument. For cross-company deal queries (every deal matching some criteria, not anchored to this company), use get_deals instead (filterable by indication (hierarchy-expanded, includes broad areas like oncology), company, role, deal type, value, and date). Each financing round and deal carries a value_status ("confirmed" or "unconfirmed"); when "unconfirmed", the USD value is null and the amount is not in Maven's data, not necessarily missing in the world. Even when you do not opt in, the default company response includes related_previews.financings, related_previews.investments, and related_previews.deals teasers (count plus a few sample rows) plus a follow-up hint, so you can see at a glance whether funding or deal activity exists before requesting the full per-row detail. When you request specific aspects, only those sections are returned; a hint lists the other available aspects so you can fetch them in a follow-up call if needed. Aspects that do not apply to the given entity_type (for example financials on a drug, or indication_phases on a company) are ignored rather than causing an error, and a hint reports which were dropped. Establishes the baseline fact pattern for one entity. search_documents and read_document provide source-backed evidence for the claims in this profile. If you only need canonical single-entity lookup without the full profile, use match_entity. If entity resolution fails, the error will suggest alternatives. For indication-level competitive landscapes, use research_landscape instead.research_landscape
ChatGPTindication -- "NSCLC", "non-small cell lung cancer", "triple-negative breast cancer" all work. If you have additional disambiguating context, pass it in context rather than packing it into indication. This tool takes ONE indication, itself plus its descendants -- so for a broad commercial area ('immunology', 'heme onc', 'cardiometabolic'), pass a broad umbrella indication (e.g. 'Immune System Diseases', 'Neoplasms') and it expands to the descendants, then drill into specific indications here. Programs are ranked by indication-specific phase so approved and late-stage assets surface first. The response is exact and paginated over the full matching program set, not a sampled baseline. By default, returns a compact matched-context response: each program foregrounds the indication, phase, and a capped sample of trials that explain why it matched the query. Set supporting_trial_limit to tune the trial sample size, or set detail_level='full' for heavier debugging fields such as all companies and all indication-scoped trials. Use group_by to control how the returned page is organized: - "phase" (default): programs grouped by development phase - "company": programs grouped by sponsor - "mechanism": programs grouped by mechanism of action. When a mechanisms filter is present, programs are grouped by the mechanism values that matched that filter; other co-mechanisms remain on the program as context. Without a mechanism filter, multi-mechanism programs appear in each mechanism group, so mechanism group counts are assignment counts and may sum above total_programs. Filters (defaults surface active, non-biosimilar competitive programs): - drug_phases: filter on the drug's highest phase for this indication (e.g. ["Phase III", "Phase II"]). Accepts Roman (Phase III) or Arabic (Phase 3). A drug whose highest phase for this indication is "Marketed" is NOT returned by a ["Phase III"] request, even though it ran Phase III programs to get there. None means all phases. - development_status: "active" (default), "active_plus_ndr", or "all". - exclude_biosimilars: default true. Set false to include biosimilars. - mechanisms: filter to specific MoAs (e.g. ["PD-1"]). - offset: absolute result offset for pagination. Use has_more + more_via to continue; do not increment offset by the requested limit because large pages may be truncated to an effective_limit. Each compact program returns: drug_id, drug_name, company_names, mechanisms, indication, phase, status, supporting_trials, and supporting_trial_count. phase/matched_indication_phase come from indication-specific program rows when available, falling back to indication-scoped trial phases. status is retained for compatibility as the first returned supporting-trial status; use drug_global_status, drug_sub_global_status, and supporting_trial_statuses for explicit provenance. When a precomputed relevance score exists for the program in this indication, it also carries a relevance block: {score, confidence, signals:{stage, position, activity, differentiation}, drivers, caveats}. drivers names the strongest one or two signals in plain words; caveats names data-quality flags (e.g. crowded class, readout date is an estimate). Programs without a score omit the block. This score is surfaced to users as 'Prominence'. It is a within-indication competitive read: how prominent a program is among the other programs competing in this same indication, blending its development stage, clinical position, recent development activity, and mechanistic differentiation, normalized against that indication's field. It is NOT a prediction of clinical success, approval likelihood, or commercial quality, and it is comparable only within one indication, never …search_documents
ChatGPTentity_name to the raw entity name and put extra identifying text such as sponsor/company in context. This tool is for generic document discovery and evidence lookup across Maven's corpus. When host web fetches are blocked, use this as the primary fallback path. Maven often has indexed copies or near-equivalent documents that you can then read with read_document. search_mode controls retrieval strategy: 'auto' picks the best Maven retrieval path for the query; 'semantic' uses Maven's internal semantic document search only; 'keyword_strict' uses internal lexical search requiring resolved entity terms when available. To read a document's full content after finding it, use read_document with the doc_id. Pagination: count is the number of results in this response. pagination.total counts the retrieved window, which is a lower bound on the matching corpus unless pagination.total_is_exact is true, meaning retrieval reached the end of the matches. Page while pagination.has_more is true, following more_via. A retrieval_depth_reached hint means retrieval stopped at its own ceiling, so narrowing the query surfaces more than paging deeper will.search_entities
ChatGPThints to guide the next step. Results include typed IDs that can be passed to research_entity, read_document, get_financials, or get_financings for deeper data. Search results are discovery candidates; search_documents, read_document, and research_entity provide source-backed detail on any candidate. For drug searches, when Prominence scoring is available the results are ordered Prominence-first (the highest-signal programs lead, so the first page is not polluted by less relevant matches), and each scored drug carries a relevance block: {score, confidence, signals, drivers, caveats}. Prominence (surfaced to users by that name) is a within-indication competitive read: how prominent a program is among the others competing in the same indication, blending development stage, clinical position, recent development activity, and mechanistic differentiation, normalized against that indication's field. It is NOT a prediction of clinical success, approval likelihood, or commercial quality, and is comparable only within one indication, never across indications. No candidate is dropped for low Prominence; treat a missing block as 'no precomputed score', not 'low prominence'. Note on company discovery: results match the filters but are NOT relevance-ranked; broad company queries return large alphabetical candidate sets (match_score may be null) where well-known names can be absent from page 1. For "top companies" or ranking questions, use aggregate_records instead (e.g. entity_type='drug', query='drugs grouped by company' plus your filters) and treat search_entities output as an unranked candidate pool.aggregate_records
Claudefetch_related
Claudeget_artifact
Claudeget_deals
Claudeget_financials
Claudeget_financings
Claudeget_monitors
Claudeget_recent_events
Claudelist_artifacts
Claudematch_entity
Clauderead_document
Clauderesearch_bio_evidence
Clauderesearch_entity
Clauderesearch_landscape
Claudesearch_documents
Claudesearch_entities
ClaudeLinks
Similar Apps
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/maven-bio"><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
[](https://mcp-app-store.keenethics-labs.com/app/maven-bio)






