Whatever your customers send you — CSV, Excel, SQL dumps, JSON, XML — AdaptivMapr maps it to your schema and hands back validated rows. Typos, German or Spanish column names, free-text values like "hemoglobin a1c" landing on the right code: all handled, so your downstream code doesn't have to.
Three surfaces
The same engine powers all three: a REST API for your product, a hosted Workbench for your team, and an MCP server for your AI agent. One contract, one validator library, one audit trail — whichever surface you start with.
POST /v1/uploadsREST API + SDKs.
Embed the transform call into your import flow. One API call: file in, structured rows out. Webhook delivery (HMAC-signed), inline response for small payloads, or 20+ built-in validators on the way through. Stateless auth so there's no key lookup latency.
app.adaptivmapr.comThe Workbench.
Drag and drop a file, watch the mapping happen live, edit anything that looks wrong, then export to CSV or push to a webhook. Built for the moments when you don't want to write code — onboarding a new customer, debugging a problem feed, prototyping a new schema.
npx -y @adaptivmapr/mcp-serverMCP server.
Your Cursor or Claude agent picks up data adaptation as a first-class tool. Ask it to import a CSV, map to a target schema, and ship the rows somewhere — and it does. Schema-only mode is free and needs no key; full-data tools require a paid plan.
Mapping engine
Most column mappings are easy: an exact match, a known synonym, a tiny typo, a translation. AdaptivMapr handles those for free. The LLM only fires on the genuinely ambiguous columns — which means predictable bills, not surprise invoices when a customer ships you a 200-column export.
When a column has been mapped the same way enough times, AdaptivMapr auto-accepts it the next time it sees that header. The headers your customers send most often stop costing you anything to process.
Exact and synonym matching against multilingual hints (DE / FR / IT / EN / ES). Vorname, prénom, nombre all land on first_name without a model call. Your Swiss customer's roster maps without paying for AI.
Typo tolerance, reordered words, weird punctuation. Frist Name still resolves to first_name. The columns your customers fat-finger don't need an AI escalation; they just work.
Embedding similarity catches the multilingual long tail. Birthday lands on date_of_birth. Headers and field embeddings are cached per process, so the same column never gets re-embedded.
Only fires when the four cheap layers all miss. The model is constrained to your allowed column set, so it can't invent a field. Bring your own provider key for 50% off metered usage.
AUTO_ACCEPT_RULES = [{ minN: 100, minRatio: 0.95 }, { minN: 20, minRatio: 1.00 }]Thresholds production-tested across employee rosters, patient lists, supplier catalogs, stock movements, timesheets, and expense reports. Unchanged since GA.
Start fast
Twenty-three production-tested schemas across four domains. Each ships with multilingual hints, field-level validators, and FHIR mappings where applicable. Or skip the catalogue and tell AdaptivMapr what you need in plain English.
Users, transactions, addresses. The shapes most apps start with.
Coming soon: Payroll, Logistics, HR.
Patient demographics, lab results, drug formularies. FHIR-compatible out of the box.
Coming soon: Payroll, Logistics, HR.
Leads, contacts, opportunities, accounts, activities. Drop in next to Salesforce or HubSpot.
Coming soon: Payroll, Logistics, HR.
Orders, products, customers, inventory, returns. Built for storefront imports.
Coming soon: Payroll, Logistics, HR.
Paste your headers, describe what you want in English, or upload a sample — AdaptivMapr will generate a target schema and map your data to it in one shot. Save it to your workspace and the next import just uses it.
users_v1transactions_v1addresses_v1employee_roster_v1patient_demographics_v1lab_result_catalog_v1drug_formulary_v1claims_line_items_v1appointment_log_v1insurance_contracts_v1supplier_inventory_v1provider_directory_v1lab_results_v1leads_v1contacts_v1opportunities_v1accounts_v1activities_v1orders_v1products_v1customers_v1inventory_v1returns_v1Built in
Format parsers. Validators. Multilingual matching. Audit logs. Webhook delivery with HMAC signing. All the unglamorous infrastructure your team would spend two quarters building, already shipping in production.
CSV, TSV, Excel, JSON, XML, SQL. We figure out the format, the headers (even when they sit on row 4), and the encoding. You point us at the file.
Most mappings happen deterministically — matching, typo tolerance, semantic synonyms. The LLM only fires on the genuinely ambiguous columns. Your costs don't scale with your data volume the way a naïve LLM pipeline does.
AdaptivMapr doesn't persist row content. API keys are self-signed tokens. Schema-only mode keeps PHI and PII out of the pipe entirely. Your security team approves us without a DPA cycle.
Every transform is hash-chained: which source column landed in which target field, with what confidence, via which layer. Auditors get one URL and see the chain.
Templates ship with hints in five languages. Geburtsdatum, date de naissance, fecha de nacimiento, data di nascita — all map to date_of_birth without firing the LLM.
Inline JSON response, signed webhook, or direct-to-Postgres from a self-hosted worker. Pick what fits your stack.
MCP server
Your AI agent gets data adaptation as a first-class tool. Ask it to import a CSV, map to your users schema, and send the rows to your webhook — and you get structured rows back. The free tier registers without an API key; gated tools require a paid plan.
adaptivmapr.list_templatesBrowse the template catalogue from inside your agent.adaptivmapr.template_schemaRead the target schema for any named template — columns, types, validators.adaptivmapr.match_headersAsk the agent to map your CSV headers to a schema. Returns ranked mappings with confidence and alternatives.adaptivmapr.validate_rowCheck a single row against a schema. Nothing leaves your machine that you didn't paste in.adaptivmapr.csv_previewLocal CSV peek so the agent knows what it's looking at. No network call.adaptivmapr.match_full_fileFull-file mapping with LLM cleanup on the long tail. Requires a paid key.adaptivmapr.commit_to_webhookFinalise an import and stream the validated rows to your webhook.adaptivmapr.commit_to_databaseWrite directly to your Postgres via the self-hosted worker.{
"mcpServers": {
"adaptivmapr": {
"command": "npx",
"args": ["-y", "@adaptivmapr/mcp-server"],
"env": {
"ADAPTIVMAPR_API_KEY": "am_live_...",
"PHI_GATEWAY_API_KEY": "phi_live_..."
}
}
}
}Why AdaptivMapr
Data adaptation isn't a tool decision. It's an infrastructure decision — one that touches your auth, your billing, your security review, and every customer-onboarding meeting for the next two years. Here's why teams pick AdaptivMapr.
Your team is paying engineers to maintain CSV import code that breaks every time a customer ships a new variant. Different delimiter, headers on row 4, German column names, dates in the wrong format. AdaptivMapr absorbs that drift so you can route the engineers at something that actually moves your roadmap.
Real-world feeds carry typos, abbreviations, translated headers, free-text values where you expected codes, and entire columns nobody asked for. AdaptivMapr reconciles the source to your target without code changes — and surfaces the few rows that genuinely need a human glance.
The mapping engine and the templates are learned from millions of real business imports — employee rosters, patient lists, supplier catalogs, stock movements, expense reports. The defaults reflect what real customers actually send, not what an engineer hoped they'd send.