Edge Parser

AI document memory · MCP-first

Every invoice, receipt and contract your business gets — remembered, searchable, ready for AI.

Edge Parser captures documents from every source you already use — email, upload, cloud storage — and turns them into queryable memory. Your team finds anything in seconds. Your AI agents — Claude, ChatGPT, Cursor, or your own — reach in directly. Reconciliation, lookup, audit — all from one place.

Request access See how it works
From CloudMailin · Email · Upload · Cloud storage · MCP server included · Your data, your bucket

The problem

Your business documents are everywhere — and nowhere.

Invoices arrive by email. Receipts get downloaded to a folder. Statements live in three different supplier portals. Contracts sit in shared drives with names like final_v2_REAL.pdf. When you need one — to reconcile a payment, settle a query, or answer an audit — you spend twenty minutes hunting.

Modern AI agents could do this work for you in seconds. They can't — because the documents aren't anywhere they can reach. AI features bolted onto storage tools (Drive AI, Box AI, Notion AI) treat documents as a search-after-the-fact afterthought. Edge Parser is built the other way around: AI-first, MCP-native, with documents as the queryable substrate of your business.

How it works

Three primitives. One memory.

Edge Parser exposes the same three operations to people, your apps, and AI agents — over a clean MCP interface and a plain HTTP API.

Search

Find any document by content, sender, date, type, or structural metadata. Full-text + jsonb queries underneath, one tool surface on top.

search(query: "Ubiquiti invoices since March")

Show

Open a document with its full context — the original email, sibling attachments, summary, and metadata in one response.

show(id: "…", include: ["siblings", "raw"])

Edit

Annotate, correct, retype. Summaries are mutable — humans and agents can both keep memory accurate over time.

edit(id: "…", summary: "Hosting · Q2 2026")

First use case

Invoice & payment reconciliation, automatic.

Match every bank transaction to the invoice or receipt that justifies it — without your bookkeeper lifting a finger. Edge Parser indexes every supplier email and uploaded receipt; your AI agent does the matching against Xero, QuickBooks, NetSuite, or whichever accounting tool you already use.

  • Forward your supplier emails. Drop in PDFs. Connect your cloud storage. Pick any combination.
  • Search by vendor, amount, date, or any natural-language query.
  • AI agents reconcile transactions to documents over MCP — no plugin glue.
  • Audit-ready: every match links back to the original document and source email.
// Your accounting agent, talking to Edge Parser over MCP
User:  Reconcile last month's bank feed.
Agent: → search(query: "invoice", from: "2026-04-01", to: "2026-04-30")
       Found 47 invoices totalling £18,431.

Agent: → matches transactions to documents
       42 matched automatically.
       3 partial matches (vendor + amount, dates off).
       2 unmatched — likely missing supplier emails.

Agent: → show(id: "01j…") for each unmatched txn
       Posted to Acme Accounts with attached receipts.

Result: 5 hours of bookkeeper work done in 90 seconds.

Reconciliation is the launch use case. The same primitives serve contract review, support context, supplier lookup, and any other business document workflow.

See the full reconciliation walkthrough

How MCPs compose

Edge Parser is one MCP. Your agent uses many.

Edge Parser handles the document side of the reconciliation: searching, showing, and updating records. The accounting side — reading bank transactions, posting explanations, attaching receipts — comes from your accounting MCP. Your AI agent composes the two.

Edge Parser MCP

Documents

search, show, edit your invoices, receipts, contracts.

+

Your accounting MCP

Books

Xero, QuickBooks, NetSuite, your bank feed — whichever you use.

=

Your agent

Reconciles

matches transactions to documents, posts explanations, attaches receipts.

Same pattern for any workflow: Edge Parser + your CRM MCP for support context, Edge Parser + your contract MCP for renewals, Edge Parser + anything else MCP-aware. Compose, don't lock in.

Every format

PDFs, spreadsheets, scans — all in one archive.

Whatever your suppliers, banks and customers send you, Edge Parser stores it, indexes what it can, and makes the result queryable. No format gets dropped on the floor.

Text-heavy documents

PDF Word .eml HTML Markdown .txt

Full text extracted, AI-summarised, and ranked by relevance. Search any phrase, sender, or topic and the right document surfaces.

Spreadsheets & data

CSV Excel .xlsx OFX QFX JSON

Stored with structure preserved — rows and columns stay queryable, not flattened into prose. Bank exports, supplier lists, financial reports, all in shape.

Images & scans

JPG PNG HEIC TIFF WEBP scan-PDFs

Stored and thumbnailed for fast browsing. Vision and OCR run on demand — turn a phone photo of a receipt into searchable text and structured fields.

Anything else — archives, ZIPs, anything you upload — gets stored alongside its ingestion context (sender, timestamp, batch) so it's still findable, even if Edge Parser can't read inside the file itself yet.

A workspace for humans, too

When AI isn't the right tool, the UI is.

For browsing, audit, manual review, or just answering "did that arrive?" — Edge Parser comes with a polished, fast UI on desktop and mobile. Same data, same primitives, two surfaces.

edgeparser.cloudmailin.com / Inbox
Edge Parser desktop UI showing the master-detail browser: a left-side list of ingestions and the selected document on the right.
Master-detail browser: list on the left, document on the right. Filters, search, and AI Q&A drawer one click away.
Edge Parser mobile UI: ingestion list with bottom navigation and search.
Touch-optimised mobile UI. Bottom nav, swipe-back navigation, the whole archive in your pocket.

Master-detail browser

List on the left, document on the right. Keyboard navigation throughout.

Drag-drop upload

Pull files in from anywhere on your desktop. Multi-file batches stay grouped.

Live updates

New documents appear in the list the moment they arrive — no refresh needed.

Ask the archive

Natural-language Q&A drawer right inside the UI. Same answers your agents get.

Ask the archive

Stop hunting. Just ask.

Edge Parser ships with natural-language Q&A built in — not bolted on. Your team gets it as a panel inside the UI; your AI agents get it over MCP. Same brain, same documents, same answers, with every claim grounded in the source material.

Ask the archive

Live
How much did we spend on AI tools last quarter, and which was the biggest?

Across Q1 2026 (Jan–Mar), spend on AI tools came to £3,847 across four suppliers:

  • Anthropic (Claude API + subs)£2,104
  • OpenAI£978
  • Cursor£540
  • ElevenLabs£225

Anthropic was the biggest — mostly API token spend on the production agent (£1,712), with the rest in team Claude subscriptions.

Sources

Anthropic invoice 0421 OpenAI receipt Mar Cursor Q1 statements + 5 more
Ask a follow-up…

You can also ask…

  • "What was the last invoice from Hetzner?"
  • "Total billed to AWS across all accounts last quarter"
  • "When does our cloud storage contract renew?"
  • "Find the email where Acme Corp confirmed the discount"
  • "Which subscriptions auto-renewed in May?"
  • "Did we ever get the credit note for invoice 24142?"

Cited

Every answer

Streamed

As it thinks

Threaded

Follow-ups land

Bring your own AI

Works with Claude, ChatGPT, Cursor — or whatever you build next.

Edge Parser speaks MCP (Model Context Protocol) — the open standard for plugging AI assistants into external data. Connect Edge Parser once as an MCP server, and any compatible AI can search, show, and edit your documents directly. No bespoke integrations to maintain. No data pre-loaded into someone else's prompt.

Anthropic

Claude

Native MCP support in Claude Desktop, Claude Code, and the API. The canonical home for MCP integrations.

OpenAI

ChatGPT

MCP-compatible connectors in ChatGPT and the OpenAI API. Configure once, use anywhere.

IDEs & coding

Cursor, Windsurf…

Every MCP-aware coding assistant — Cursor, Windsurf, Continue, Cline. Your agent gets document context alongside code context.

Your build

Custom agents

LangChain, LlamaIndex, your own Python script, anything that can speak MCP. Plain HTTP API also available.

New AI tool tomorrow that speaks MCP? Edge Parser will work with it on day one. That's the point of the standard.

For developers & AI engineers

MCP-first. Two lines of config.

Edge Parser is built around the Model Context Protocol from day one — not retrofitted. Point Claude, ChatGPT, your custom agent, or anything else MCP-aware at the endpoint and it has search, show, and edit available immediately.

  • Streamable HTTP MCP transport over bearer-token auth.
  • Postgres-backed full-text + structured jsonb metadata queries.
  • Cloud storage for raw artefacts; signed URLs on demand.
  • Open API for custom integrations beyond MCP.
// .mcp.json — your agent config
{
  "mcpServers": {
    "edge-parser": {
      "url": "https://your-tenant.edge-parser.cloudmailin.com/mcp",
      "transport": "http",
      "headers": {
        "Authorization": "Bearer ep_live_…"
      }
    }
  }
}

That's it. search, show, and edit are now available to your agent.

Security & data sovereignty

Your data, your choice — and never anyone else's product.

We sell document infrastructure. We don't sell your documents. Pick the deployment model that fits your compliance posture — from fully managed to fully on your own infrastructure.

Hosted

Default

Fully managed for you. Documents in AWS S3, index in our Postgres database. Single-tenant per customer. Fastest path to production.

Documents
AWS S3, region of your choice
Index
Our database
Best for
Most SMB and team setups

Bring your own bucket

We hold the metadata index. Your raw documents go straight into your AWS S3 bucket via cross-account IAM role — bytes never enter our infrastructure.

Documents
Your AWS S3 bucket, your region
Index
Our database
Best for
Regulated industries, larger orgs

Self-hosted

Run the whole stack on your infrastructure. Your database + any S3-compatible object store + the Edge Parser binary. No data ever leaves your network.

Documents
Your S3 (or any S3-compatible)
Index
Your database
Best for
Air-gapped, on-prem, max control

What we commit to

  • No model training on your data. Your documents are never used to train, fine-tune, or improve any AI model — ours or anyone else's.
  • No selling, no sharing. We don't sell, license, or hand your documents to third parties. Ever.
  • Standard formats, no lock-in. Documents stay in their original form — PDFs as PDFs, emails as raw .eml. Bulk export the entire archive whenever you want.
  • Encrypted in transit and at rest. TLS for every request; AWS S3 SSE for objects; Postgres at-rest encryption.
  • GDPR compliant. Pick the AWS region that fits your data-residency requirements — UK, EU, US, or anywhere else AWS operates.
  • Delete anything, anytime. One-click export and full account-level deletion. In BYO mode, we never had the bytes to begin with.

Built on the same infrastructure that powers CloudMailin — trusted with billions of business emails since 2010.

Private beta

Bring your business documents into AI memory.

Edge Parser is in private beta. Tell us what you'd like your AI agent to know about and we'll get you in.

No marketing list. We'll reply personally.