Extract document data through MCP with Airparser
Connect your AI agent to Airparser to upload documents, use saved extraction schemas, and retrieve structured JSON. Work with your existing inboxes and parsing workflows from a compatible MCP client.
Part of your AI stack
Use Airparser as the system that structures document data, while AI agents reason about workflows, exceptions, and next actions.
Work with real pipelines
Instead of one-off file analysis, agents can inspect inboxes, review parsed history, update schemas, and test post-processing.
Repeatable output
Keep the benefits of structured extraction: saved schemas, document history, and post-processing logic that agents can use and improve over time.
Why Airparser MCP is different
Compared with a basic manual upload-and-prompt workflow, Airparser MCP gives an agent a saved inbox, extraction schema, document history, and reusable post-processing.
A basic manual upload-and-prompt workflow
- One document at a time
- No saved extraction schema
- No document history or inbox context
- No repeatable post-processing layer
Using Airparser MCP
- Agents can work with inboxes, schemas, and parsed documents
- Structured, repeatable output stays at the center
- Schema and post-processing changes can be tested and refined
- Parsed data becomes usable across broader AI-driven workflows
What users can do with Airparser MCP
The big advantage is not just extraction. It is giving AI agents the context and controls they need to improve document workflows.
Classify and route incoming documents
Separate invoices, resumes, purchase orders, and shipping documents before they enter the right parsing workflow.
Backfill historical emails and files
Import old emails, attachments, and documents from other apps to rebuild pipelines faster and recover structured data from past operations.
Generate and update extraction schemas
When layouts change, an agent can help adapt fields and schema structure instead of making your team reconfigure everything by hand.
Write and test post-processing code
Let agents help clean, normalize, enrich, and reshape parsed data before it is exported to the next system.
What actually happens where
Checked against the current Airparser MCP server (17 tools) so nothing here is a guess.
| View an inbox and its schema, or get the parsed JSON for a document | Airparser MCP — list_inboxes, get_inbox, get_extraction_schema, get_document, list_documents (all read-only) |
| Create an inbox, upload and process a document, edit the schema, or write post-processing code | Separate tools in Airparser MCP that create or change something (create_inbox, upload_document_sync, update_extraction_schema, save_postprocessing_code, and others) — not the same as the read-only tools above |
| Pull an email from Gmail or write results into another app | A different connector or tool configured in your agent. Airparser MCP does not reach other apps on its own. |
| Hold a document for review and approve the result | Check the review setting on the inbox itself — Airparser MCP has no dedicated approval tool today. |
Requires an Airparser account and a compatible MCP client. Tool names and availability follow your current Airparser MCP server version — check your client's tool list before relying on a specific one.
Real example
PDF → saved schema → JSON, the way an agent fetches it
The schema below belongs to the same demo inbox used on the PDF table parser page — reachable today through Airparser MCP with the steps underneath. See the note under the results for exactly where this specific example came from.
What an agent does, step by step
- 1Connect a compatible MCP client (Claude, ChatGPT, or another MCP-aware agent) to your Airparser account.
- 2List inboxes and open the one with the right schema (list_inboxes, get_inbox).
- 3Read the current extraction schema for that inbox (get_extraction_schema).
- 4Upload the document and wait for parsing to finish (upload_document_sync).
- 5Read back the parsed JSON (get_document) — or list_documents to find it again later.
Example prompt
Use the demo inbox and its existing extraction schema. Upload this sample PDF, wait for parsing to finish, and return the extracted JSON. Do not change the inbox schema or post-processing settings. Report any parsing error instead of inventing missing values.
Extraction schema for this inbox
6 scalar fields, plus a repeating "items" list with 5 columns per row — object + line items, the shape most buyer questions ask about.
Parsed line items
| item_code | description | quantity | unit_price | line_total |
|---|---|---|---|---|
| WH-1001 | Steel bolt M8 x 40 mm, zinc plated (box of 100) | 12 | 14.50 | 174.00 |
| WH-1002 | Hex nut M8, zinc plated (box of 100) | 12 | 9.80 | 117.60 |
| WH-1015 | Flat washer M8, stainless (box of 200) | 8 | 11.25 | 90.00 |
| WH-1120 | Cable tie 300 mm, black (pack of 100) | 25 | 4.60 | 115.00 |
| WH-1121 | Cable tie 200 mm, natural (pack of 100) | 20 | 3.75 | 75.00 |
| WH-1210 | Work gloves, nitrile coated, size L (pair) | 40 | 2.90 | 116.00 |
| WH-1211 | Safety glasses, clear lens, anti-fog coating | 30 | 3.40 | 102.00 |
| WH-1305 | Duct tape 48 mm x 50 m, silver | 18 | 5.20 | 93.60 |
| WH-1306 | Masking tape 24 mm x 50 m | 24 | 2.35 | 56.40 |
| WH-1410 | LED work light, rechargeable, 10 W, with magnetic base | 6 | 27.90 | 167.40 |
| WH-1411 | Extension cord 5 m, 3 x 1.5 mm2, indoor use | 10 | 12.60 | 126.00 |
| WH-1502 | Pallet wrap film 500 mm x 300 m, clear | 15 | 8.40 | 126.00 |
| WH-1503 | Packing tape 48 mm x 66 m, brown | 36 | 1.95 | 70.20 |
| WH-1620 | Storage bin 600 x 400 x 220 mm, stackable, with lid | 14 | 9.15 | 128.10 |
| WH-1621 | Label roll 100 x 150 mm, direct thermal (roll of 500) | 10 | 13.70 | 137.00 |
| WH-1700 | Hand truck, folding, 70 kg capacity | 2 | 64.00 | 128.00 |
This specific result was captured on 2026-09-19 through the app (Vision engine), on this inbox's schema, and checked row by row against the source PDF at the time. During this update we also ran the same file through Airparser MCP's upload_document_sync twice: one run returned 15 of 16 rows, another 9 of 16, with the same inbox settings. We haven't identified the cause yet, so treat the schema and steps above as accurate, but don't treat this 16/16 result as a guarantee that a fresh run reproduces it. Layout, scan quality, and document length also affect results generally — test on your own documents before relying on this for production volume.
Want to see this run against your own document?
Start for freeExample MCP workflows
Airparser MCP is especially useful when documents are only one part of a larger AI-assisted process.
Operations teams
- “Review our inboxes and tell me which schemas likely need an update based on the latest documents.”
- “Pull the latest parsed invoices, flag missing totals, and test a post-processing fix.”
- “Backfill historical attachments from our mailbox into Airparser so we can rebuild reporting.”
AI product builders
- “Create a schema for these shipping documents and test it on the latest uploads.”
- “Fetch the parsed JSON and send the result into the next agent step for validation and routing.”
- “Use Airparser for extraction, then let the agent update spreadsheets, send emails, or trigger follow-up actions.”
Questions buyers ask about Airparser MCP
How is this different from uploading a PDF into a chat?
A basic manual upload-and-prompt workflow reads one document, once, with nothing saved. Airparser MCP works with a saved inbox instead: the same extraction schema, a document history, and post-processing logic your agent can reuse on the next document — not just this one.
Can I extract my own fields, objects, and line items?
Yes. Airparser's extraction schema supports scalar fields, nested objects, repeating lists (line items), and enums — the example on this page uses a list field for line items. Results depend on how consistent the document layout is.
How schemas workDo I need to write code?
No. Connecting a supported MCP client and using an existing inbox and schema needs no code. Building custom automation around the results — routing the JSON into other systems on your own logic — is a separate, optional step that may need code or a no-code tool like Zapier, Make, or n8n.
What can an agent actually change in my account?
Read tools retrieve existing data — an inbox, its schema, or a parsed document. Write tools can create an inbox, upload and process a new document, change a schema, or edit post-processing code — see the table above for which is which. Separate tool names don't stop an agent from calling a write tool on their own; whether your MCP client asks for confirmation before running one depends on that client, not on Airparser MCP.
Can the agent read my Gmail directly?
Not through Airparser MCP by itself. Airparser can receive documents by email forwarding or through a configured IMAP/Gmail import on the inbox; having an agent read Gmail as a separate action needs its own connector or integration in your agent, outside of Airparser MCP.
How do I handle a result that needs review before it's used?
Turn on review for the inbox and choose how results are delivered — for example, wait for approval before a connected integration receives the data. Airparser MCP itself has no separate approval tool today; review happens in the app or through your configured delivery settings.
Human-in-the-loop reviewConnect Airparser MCP to your AI workflow
Start with a real use case: inspect an inbox, review parsed documents, update a schema, or test a post-processing step.
