How to Use the Lido MCP to Extract Data from Veterinary Intake Forms

A two-DVM practice ships somewhere between 30 and 60 intake forms a week — new client + new patient flows, annual wellness, surgical pre-ops. Each form is structured-ish data (owner, address, phone, pet name, breed, age, weight, vaccination history, allergy flags) plus narrative ("anything we should know"). Each gets typed into the PIMS by a CSR or tech, line by line.

Lido is the modern spreadsheet + OCR tool that handles document-to-structured-data conversion. Its MCP server wraps the same flow for agentic use: a form arrives in the practice's intake folder, Lido extracts the structured fields, Claude chains the result into client comms in your practice voice.

💡 The stack. Lido MCP handles the OCR + structured extraction. The AI Cowork Vault Bundle provides the agentic skill layer (no vet-specific vault yet — coming if Tier 2 light products validate). For now, raw Claude + bundle skills like /client-communication work fine. $49 one-time for all 7 vaults if you want the cross-profession workflow toolkit.

What the Lido MCP actually does

Lido's product is a hybrid spreadsheet + document-extraction tool. The MCP exposes:

• extract_form — given a document (PDF or image), extract structured fields against a template you define
• extract_table — given a document with tabular data, return rows
• anomaly_flag — flag fields that don't match the expected pattern (missing required fields, suspicious values)

For veterinary intake, you'd define a template once: owner name, contact info, pet name, breed, sex, DOB, weight, current meds, allergies, prior vet, vaccination records, emergency contact. Lido learns the form layout; future extractions hit it cleanly.

Pre-flight

• Claude Cowork (desktop) or Claude Code
• A Lido account — free tier covers ~100 doc extractions/month; paid for higher volume
• A scanning workflow that lands intake forms in a folder Claude can read (Drive, Dropbox, or local)
• A defined Lido template matching your intake form

Configure the MCP server

Restart Claude. The Lido server appears in connectors. You'll define the template once in Lido's UI; the MCP references it by name.

Run an intake extraction

When an intake form lands in your folder:

Lido returns the parsed structured data:

The anomaly flag is the useful part. Lido sees "NONE NOTED" but flags that the free-text section contradicts. A tech reviews; the allergy "maybe chicken" gets recorded properly.

Chain into PIMS-ready format

Most PIMS (Cornerstone, Avimark, ezyVet) have a structured import format. Have Claude convert:

You get a clean import payload + a one-line note for the tech.

Chain into the new-client welcome email

Out comes the welcome email. Pets-by-name (Buddy, not "your pet"). Practice voice. Appointment logistics. State-required intake disclosure language.

Chain into the post-visit follow-up

After the wellness visit, the same context drives the follow-up:

Five-minute follow-up email becomes a 30-second drafting task. The DVM dictates the medical content; the vault renders the owner-facing prose.

What this doesn't do

It doesn't replace your PIMS. The PIMS is still the system of record for medical history.

It doesn't auto-import to PIMS. Most PIMS lack good API surfaces; the structured output is for the tech to paste into the PIMS, not for autoposting.

It doesn't diagnose. Lido extracts text. Claude organizes prose. The DVM diagnoses.

Common failure modes

• Handwritten fields extract poorly. This is unavoidable on intake forms with handwriting boxes. Redesign forms with checkboxes where possible; route handwritten fields to humans.
• Vaccination record images aren't OCR'd reliably. They're often photos of paper records, with poor lighting. Route those to tech review; don't try to extract them automatically.
• Allergy field contradictions. Common pattern: client checks "no allergies" but writes a note in free text. Lido flags these reliably; make tech review of flagged items part of the SOP.
• Two pets on one intake form. Some forms allow multiple-pet entries. Define a Lido template for the multi-pet case; running the single-pet template on a multi-pet form will extract only the first.

When you'd skip this stack

• Under 15 intake forms/week. Manual entry is faster than the setup.
• All-digital intake already. If your clients fill the intake form on a tablet in the lobby (with structured fields) and it auto-imports to your PIMS, you don't need the OCR step. Use the chain-into-welcome / chain-into-follow-up parts only.
• Mobile / house-call practice. Different workflow; intake happens in the home with paper. Manual entry is the norm.

Sources

• Lido: Developer documentation
• Anthropic: Model Context Protocol (MCP) introduction
• AVMA: Recordkeeping guidance