Questions the person actually doing the work asks.

Yes, via a short forwarding setup. SideQuest's email integration uses the official Gmail API, so the inbox the connector reads has to be a Gmail mailbox. But that mailbox is invisible to your buyers — you create a free dedicated Gmail just for SideQuest, forward your real [email protected] mailbox to it, label the forwarded mail, and configure Gmail's "Send mail as" so replies still leave from your real domain. Your buyers never see the Gmail in the middle.

Two step-by-step guides, pick the one that fits your setup:

Microsoft 365 / Outlook setup — for teams on Office 365, Outlook.com, or Exchange.

cPanel / GoDaddy / Bluehost setup — for teams whose business email is hosted on shared web hosting.

About 10 minutes either way. A native Microsoft 365 / IMAP integration that skips the Gmail middleman is on the roadmap.

No. Google Workspace IS Gmail under the hood — SideQuest connects to it directly via the regular Gmail OAuth flow. Use the standard quick-start guide. The "forwarding" docs above are only for teams whose mail is NOT on a Gmail-compatible system.

Yes. SideQuest doesn't care whether the email comes from a buyer or from your own rep — same pipeline. Reps save a structured email template to their phone Notes once, fill in the customer + line items on the customer visit, send it to the office order desk address. SideQuest parses it like any other PO and the draft Estimate lands in QuickBooks usually within 60 seconds. Full setup, phone-friendly template, and rep tips on the field-order capture page.

Both. QBO is the default and the most-tested path. QBD runs as a Windows beta with the same connector, same Claude prompts, and same Insights reports. The only difference is the path the data takes from Claude to QuickBooks.

QBO path. The connector talks REST to Intuit's hosted API using your OAuth token. No local QuickBooks process required.

QBD path. A small local Flask bridge (qb-desktop-bridge) runs alongside QuickBooks Desktop on your Windows machine. The connector talks REST to the bridge; the bridge translates each call into QBXML over COM and ships it to QBD via the QBSDK. QuickBooks Desktop has to be open with your company file loaded for the bridge to read or write.

To switch backends, set QB_BACKEND=desktop in ~/.qb-distributor-mcp/.env (or online for QBO, the default), restart Claude Desktop, and the connector routes through the chosen path. Everything else — drafts, matching, auto-reply, Insights — works the same.

Beta program: we're running a small QBD beta. Apply here for free use up to 200 POs/month for 12 months in exchange for honest feedback while we validate the bridge against real customer installs.

QBD 2023+ and Enterprise v23+ on Windows. The bridge speaks QBXML 13.0, which targets those versions. Older QBD installs may need an older QBXML version stamp — if you're on an older release and the bridge fails, email us and we'll bump the version target.

QBD Mac is not supported (Intuit killed it). QBD Pro and Premier are supported today, but Intuit is sunsetting new subscriptions to those tiers through 2027; Enterprise is the long-term home. Multi-user mode works for reads; multi-user writes are not yet hardened, so single-user mode is the supported config for write operations.

Same install.bat as the QBO path. Step 5b of the installer drops qb-desktop-bridge/start-bridge.bat in your connector folder. The flow:

1. Open QuickBooks Desktop with your company file. 2. Double-click start-bridge.bat. A small command window stays open. 3. The first time the bridge talks to QBD, QuickBooks pops an "Integrated Application" trust dialog. Click Yes, always allow even if no user is logged in. 4. Set QB_BACKEND=desktop in your .env. 5. Quit and reopen Claude Desktop. 6. Ask Claude "list 5 items from my QuickBooks catalog" to confirm the path is live.

Yes. Install both. They do different jobs and never collide.

Intuit's official connector reads your QuickBooks for analytics and reporting. Per Intuit's own docs it's read-only: it can analyze your books, generate reports, and answer "what's my AR aging" or "what's my YTD revenue", but it cannot create invoices, record payments, or modify transactions.

SideQuest does the inbound side. It reads PO emails from Gmail, OCRs the PDFs, matches every line against your QuickBooks catalog, and writes draft Estimates back to QuickBooks for you to review and submit. It also drafts the reply email to the buyer.

Mechanically the two connectors register as separate MCP servers in Claude Desktop's config file. Ours lives under the namespace qb-distributor.*, theirs under quickbooks.*. No tool-name conflicts. Claude can call tools from both in the same conversation — ask "what's my YTD revenue" and Intuit's connector answers; ask "process this incoming PO" and SideQuest answers. Customers who run both get the analytics layer from Intuit and the order-entry layer from us, same Claude chat.

The other meaningful difference: Intuit's connector routes data through their cloud. SideQuest runs locally — customer PO content never reaches our servers. Distributors who care about customer-data privacy keep us in the stack for that reason alone.

No. Insights is a feature inside the same connector you already installed. Eight built-in reports answer questions in plain English: top SKUs, top customers, automation health, time saved, and two QuickBooks pass-throughs. Just ask Claude. See the Insights page for the full catalog.

Open Claude Desktop and ask in normal English. Try: "what are my top 10 SKUs by quantity this month?" or "which customers send us the most POs?" or "how much time did SideQuest save me last month?" Claude picks the right report tool, calls it, and synthesizes the answer. If you're not sure what's available, ask "what reports can SideQuest run?" — Claude calls list_reports for you.

No extra setup. The reports read your local SQLite stores (~/.qb-distributor-mcp/drafts.sqlite and ~/.qb-distributor-mcp/usage.sqlite) and your existing QuickBooks connection. No new credentials, no new API keys, no separate subscription. If you're on an older connector, download the latest zip.

No to both. Insights is included on every tier including the free tier. Report calls don't count against your monthly PO quota — they're not billable events. You can run reports as often as you want.

Local reports (top items, top customers, match quality, time saved, POs processed) read what the connector has actually processed since you installed it. They surface things QB doesn't track — review-flag rate, automation health, what's hitting your inbox. QuickBooks reports (report_qb_top_items, report_qb_top_customers) pull live from your QBO account, so they reflect your full sales history. Use both — they answer different questions.

Local reports reflect what SideQuest has processed since install. Your first month, the data is sparse. By month three the local reports become the most useful view of your PO automation health — they show trends QB can't see. If you want a deeper view today, use the QuickBooks pass-through reports (report_qb_top_items, report_qb_top_customers) which pull your full QBO history.

Intuit refuses localhost and IP-address redirect URIs on Production-flagged apps. The fix is to use the hosted callback at sidequestautomation.com/qb/callback. Steps:

1. Open your Intuit app at developer.intuit.com → My Apps → Keys & OAuth.
2. Under "Redirect URIs", click Add URI and paste exactly https://sidequestautomation.com/qb/callback.
3. Open ~/.qb-distributor-mcp/.env and set QB_REDIRECT_URI=https://sidequestautomation.com/qb/callback.
4. Re-run python -m qb_distributor_mcp.auth_qb. It now prints an authorization URL, you sign into Intuit, the hosted page shows the code + realm ID, you paste both back.

The callback page is pure JavaScript that reads the URL — nothing is sent to or stored by us. The authorization code is short-lived and only useful with your QuickBooks app's client secret, which never leaves your machine.

Three buckets. auto_matched_lines are lines the matcher got right on its own. flagged_for_review_lines are lines that were too low-confidence and got marked for human review. operator_assigned_lines are lines where you manually re-mapped the SKU after the fact, for example mapping a customer's part number to a fallback "Services" item. The split exists because counting a hand-fixed line as "auto-matched" was misleading. A high operator_assigned rate usually means you need more cross-reference rows for that customer's part-number convention.

The line gets flagged as "needs review" and the draft Estimate is created with that line marked. You see the line, you see the candidates SideQuest considered, you pick one or override with the correct QB SKU. SideQuest remembers your choice — add it to your cross-reference CSV and the next time that part shows up from that customer, it auto-matches.

Yes, always. Every match is editable in the draft Estimate. SideQuest never writes to QuickBooks without your explicit "submit" — you can fix any line by chatting with Claude before submitting.

Three layers. First, a fuzzy match on the SKU itself ("BR-ELB-050" vs "BR-ELB-O50" or "BR-ELB050"). Second, a description-based match ("stell pipe 3/4 sched 40" → your steel pipe SKU). Third, your uploaded cross-reference list. Confidence scores attach to every match so you can see what to trust.

The connector learns automatically. When you process a PO and a line is flagged because the buyer's part number doesn't match your catalog, you assign the right QB item by chat in Claude. The connector quietly appends a row to cross_reference.csv mapping that customer's part number to the QB item you chose. The next PO from that customer with the same part auto-matches at 0.99 confidence. No spreadsheet work, no admin step. By month three, the customer's whole convention is mapped just from doing normal work.

You can still pre-load cross-references the old way if you want a head start: give SideQuest a CSV mapping "ACME-EL34" → "BR-ELB-075-NPT" at install time and it picks up from there. Pre-loading + auto-learning compound.

Disable auto-learning with AUTO_LEARN_CROSS_REFERENCE=false in your .env if you'd rather keep cross-references manual.

A pricing variance is when the PO line item price doesn't match the QuickBooks list price for that SKU. SideQuest flags the line with the PO price next to the QB price so you can see the gap. You decide: keep the PO price (the customer's contract is right and QB is stale), or override to the QB price (the customer made a typo). Once you submit, the Estimate carries whichever price you picked.

No — and this is by design. If the buyer's PO supplies a unit price that's below your QuickBooks catalog price, SideQuest uses your catalog price on the draft Estimate and flags the line for review. The PO's offered price is recorded on the line so you can see exactly what they asked for. You can still override by editing the draft before submitting, but the default is: we never silently quote below catalog. If the PO price is at or above catalog, the PO price is kept (and the variance check still flags suspiciously high numbers). If the PO has no price, catalog is used.

Not today. v1 compares each PO line to the QuickBooks list price for that SKU. Per-customer pricing memory ("Acme always pays $4.85 for this part") is on the roadmap. If your contracts are well-modelled in QuickBooks already (via Price Levels or per-customer price overrides), the connector reads those.

Open ~/.qb-distributor-mcp/.env in a text editor and add the line SIDEQUEST_AR_FOLLOWUP=true. Save, then fully quit Claude Desktop (Cmd+Q on Mac, right-click tray then Quit on Windows), reopen, and re-run the env-injection one-liner from Step 5 of the install prompt so Claude Desktop picks up the new variable. Then ask Claude "run an AR follow-up sweep" in a fresh chat.

No. SideQuest's Gmail OAuth scope is gmail.modify, which lets us create drafts but cannot send mail. Every follow-up email lands in your Gmail Drafts folder. You open Drafts, skim each one, hit Send. We do this on purpose — you stay the last set of eyes on anything your customer reads from you.

It pulls every open Invoice from your QuickBooks file, classifies each into an aging bucket (due_soon, overdue 1-7 days, 8-30, 31-60, 61-90, 90+), groups by customer (so a buyer with three overdue invoices gets ONE consolidated email, not three), renders a tier-appropriate follow-up email per customer with the tone scaled to the worst-bucket invoice in their stack, and writes each one as a Gmail draft. A friendly check-in for 3-day-overdue. A hold notice for 90+ overdue.

No. AR follow-up is unmetered on every paid tier. The only meter on SideQuest is POs processed.

Almost always one of three things: (1) the SIDEQUEST_AR_FOLLOWUP=true line is missing or misspelled in your .env, (2) you didn't re-run the env-injection one-liner after editing .env (Claude Desktop reads from claude_desktop_config.json, not .env directly), or (3) you didn't fully quit Claude Desktop with Cmd+Q before reopening. Run the env injection again, Cmd+Q, reopen, try again.

SideQuest installs as a Model Context Protocol server on your computer. The connector runs locally inside the free Claude Desktop app on Mac or Windows. Your POs are read from your Gmail using your OAuth token. Your QuickBooks data is read using your QuickBooks OAuth token. Processing happens on your machine. We do not store your PO data, your QB data, or your customer pricing on SideQuest Automation servers.

Metering data only: a count of POs processed, the Gmail message_id (an opaque string), and the line count per PO. We use this to bill you on the paid tiers. We do not see PO content, customer names, item names, prices, or any business data.

Your data is on your own computer, not our servers — there's nothing for us to delete. To disconnect SideQuest cleanly: revoke the QuickBooks OAuth in your QB account settings, remove the Gmail label permission in your Google account, and delete the local connector folder if you want the config gone. (On Mac: ~/.qb-distributor-mcp/. On Windows: %USERPROFILE%\.qb-distributor-mcp\.) We're disconnected the same minute.

SideQuest drafts one Estimate per ship-to. If a single PO covers three warehouses, you get three drafts. Each one is clearly tagged with the ship-to so your warehouse team knows where to send what.

SideQuest notes the requested quantity, checks your QB qty_on_hand, and flags any line where the customer wants more than you have. You decide: short-ship, backorder, or substitute. SideQuest doesn't make that call for you.

SideQuest still tries. Description-only matching is built for exactly this. If SideQuest can extract item descriptions and quantities, it'll match what it can and flag what it can't. You might see lower confidence scores, more "needs review" flags, and a higher chance of edge cases. Worst case: SideQuest is a faster first-pass than you doing it from scratch.

English-only at launch. Spanish, French, and Portuguese are on the roadmap for Q4. If your customer base is mostly North American distribution, you're covered.

Today, via email forwarding — yes. Shopify sends an order confirmation email to your team for every order. Label those emails in Gmail (e.g. shopify-orders) and SideQuest reads them the same way it reads PO emails. The data is more structured than a typical scanned PO so accuracy is usually higher. WooCommerce, BigCommerce, and Amazon work the same way.

A direct Shopify API integration (skipping the email step) is on the roadmap for Q3 2026. Full integrations list.

v0.15.0 added one-shot bulk tools. Your rep logs in and asks Claude:

Process the overnight PO queue. Show me what's clean, what needs review, and what couldn't be parsed.

Claude calls process_overnight_queue() — which pulls every unread PO from your Gmail label, parses each one, matches lines against the QuickBooks catalog, and builds a draft Estimate per PO in a single server-side loop. Per-PO errors are isolated, so one image-only PDF can't derail the batch.

You get back a summary grouped three ways: auto_clean (ready to submit), needs_review (with specific reasons per draft — unmatched SKU, customer not in QB, price variance, etc.), and failed_to_parse (with the message_id so you can investigate).

Next, your rep says: "Preview the clean ones, then submit them." Claude calls bulk_submit_clean(draft_ids, dry_run=True) to show you the QB payload + computed total for each, then bulk_submit_clean(draft_ids, confirm=True) to push them live in one batch. Per-draft errors don't kill the batch — failed ones get isolated for follow-up.

For the review queue, the rep handles each one with the regular single-PO tools (assign QB item, link customer, override price). When done, run bulk_submit_clean on those too.

Run auto_label_unprocessed(). It scans recent unread inbox mail without any label filter, classifies each via the same PO/quote intent classifier the parse pipeline uses, and applies your PO label to anything that looks like a customer PO. Then process_overnight_queue picks them up on the next call.

Skipped messages are returned with their classification (intent: marketing, intent: notification, etc.) so you can confirm the classifier isn't dropping real POs.

One Claude chat, three or four tool calls total:

1. auto_label_unprocessed() — catch any inbox stragglers your Gmail rule missed
2. process_overnight_queue() — parse + match + draft every queued PO
3. report_review_queue() — see what's blocking, grouped by reason (skip if process_overnight_queue's output is enough)
4. bulk_submit_clean(auto_clean_ids, dry_run=True) — preview all the clean ones
5. bulk_submit_clean(auto_clean_ids, confirm=True) — push live

20 POs in maybe 60 seconds of Claude time, plus your review eyeballs on the needs_review queue.

report_review_queue(). Returns every draft in draft status grouped by the specific reason it's blocked: customer_not_in_qb, unmatched_sku, po_price_below_catalog, etc. Plus a clean list of drafts that ARE ready for bulk submit. Designed for: "show me what's blocking."

Three commands. Your .env, QuickBooks/Gmail tokens, and learned matching rules are all preserved across upgrades from v0.14.4 onward. The connector zip URL on the marketing site is always the latest version — there's no version-specific URL to remember.

1. Download + reinstall:

cd ~/Downloads && rm -rf sidequest-connector && curl -L -o sidequest-connector.zip https://sidequestautomation.com/sidequest-connector.zip && unzip -q sidequest-connector.zip && cd sidequest-connector && bash install-connector.sh

(When the installer prompts for a license key, press Return to keep the existing one.)

2. Re-inject the env block into Claude Desktop config:

~/.qb-distributor-mcp/venv/bin/python ~/.qb-distributor-mcp/reinject.py

This is destructive of the env block in claude_desktop_config.json — it mirrors .env exactly. If you've added env vars to the config that aren't in .env, add them to .env first or they'll be dropped.

3. Restart Claude Desktop:

osascript -e 'tell application "Claude" to quit' && sleep 3 && open -a "Claude"

That's it. To confirm which version you're on at any time:

~/.qb-distributor-mcp/venv/bin/python -c "import qb_distributor_mcp; print(qb_distributor_mcp.__version__)"

For what's new in each release, see the changelog.

Three signals: (1) the changelog page updates with every release — the top entry is always the latest, (2) the SideQuest control plane posts a release banner in the admin panel for paid customers, (3) on a tier with operator support, we email you when a version contains a fix that affects your workflow. There's no auto-update mechanism — we don't want to push code to your machine without you knowing — so the upgrade is always a deliberate three-command sequence (see the entry above).

Preserved across upgrades (lives in ~/.qb-distributor-mcp/):

• .env — license key, QB OAuth, Gmail OAuth, all custom env vars (v0.14.4+)
• google_token.json — Gmail refresh token
• drafts.sqlite — every PO you've worked on
• learned_rules.sqlite — auto-learned cross-reference mappings
• usage.sqlite — quota tracking

Replaced on every upgrade: the Python virtualenv (venv/) and the connector source code. That's where the new version's code goes. Your data and credentials are untouched.

v0.14.4+ fixes this. Pre-v0.14.4 installers rewrote ~/.qb-distributor-mcp/.env from scratch on every reinstall, keeping only your license key and dropping QB OAuth credentials, LICENSE_TIER, SIDEQUEST_AR_FOLLOWUP, and anything else you'd added. The combination with the env-injection (which mirrors .env into Claude Desktop's config, removing keys not present in .env) meant QB creds got silently wiped from BOTH places.

Starting v0.14.4 the installer preserves every existing key — only the license-key line gets rewritten. Re-mint your QB refresh token once via the Intuit OAuth Playground (see diagnose SYMPTOM C), edit your .env in TextEdit to paste it in, then run ~/.qb-distributor-mcp/venv/bin/python ~/.qb-distributor-mcp/reinject.py. Future reinstalls won't touch the QB lines.

v0.14.3 renamed it. Earlier versions used the internal slug "qb-distributor" for the JSON key in claude_desktop_config.json, which Claude Desktop displays in tool-use cards. The new name matches the product. The auto-migration runs the next time you call reinject.py — it copies your existing qb-distributor config block to sidequest-automation in one shot and removes the old key. No manual edits needed.

v0.14.3 added list_items(limit=25, search=None) for browsing your QuickBooks catalog without building a draft. Ask Claude "list 5 items from my QB catalog" or "find any items with 'brass' in the name". Substring search runs case-insensitively across SKU, name, and description. Caps at 200 results to keep responses fast. Useful for spot-checks, finding SKUs for a manual draft, or auditing what auto-match returned vs. what's actually in your catalog.

The license-key step was added in v0.13.0. If your previous install was older, the installer never prompted. Grab a key from your welcome email, or sign up at start-free.html for a free-tier key (delivered in about 30 seconds).

The installer preserves ~/.qb-distributor-mcp/.env across upgrades — but if the file didn't exist before (clean machine, deleted home dir, first install on a new computer), it creates a fresh .env with only your license key. QB and Gmail credentials don't carry over. Re-run the QuickBooks OAuth flow (Step 3 of quick-start) and the Gmail OAuth flow (Step 4) on any fresh-machine install. Your saved tokens and learned matching rules are preserved separately and don't need to be regenerated.

Auto-submit writes the draft Estimate into QuickBooks Online via the QBO API. The Estimate goes from "draft in SideQuest" to "Estimate in your QuickBooks Online file" without you clicking submit. That is the whole job.

Auto-submit does NOT:

• Send anything to your customer (no emails, no order confirmations)
• Email the order confirmation reply (that's a separate feature, the draft_reply_to_buyer tool, which DRAFTS the reply in Gmail and waits for you to click send)
• Convert the Estimate to an Invoice or trigger fulfillment
• "Enter" data into any other system

The flow when auto-submit fires: (1) PO email arrives, (2) SideQuest parses lines, (3) lines get matched against your QBO catalog, (4) the clean-gate runs (are all lines matched? no price variance? customer not over credit limit? no flagged risk?), (5) if everything passes, SideQuest calls the QBO Estimate API and writes the Estimate. If anything fails the clean-gate, the draft pauses for your review.

Auto-submit is opt-in (env var SIDEQUEST_AUTOSUBMIT=true) and available on Solo and above. Free-tier users can still write Estimates into QuickBooks by telling Claude "submit this draft to QuickBooks" once they've eyeballed it. One extra step per PO.

Yes, but only as a draft. Ask Claude to draft a reply after the QuickBooks Estimate is submitted (the tool is draft_reply_to_buyer) and SideQuest creates a Gmail draft on the original PO thread. The draft cites the QB Estimate number, the buyer's PO number, the total, and the lines. It lands in your Gmail Drafts folder where you open, review, tweak, and click send. SideQuest never sends mail on your behalf. We only request the gmail.modify OAuth scope, not gmail.send. Outlook / Microsoft 365 drafting is on the roadmap, same shape, different provider.

Yes. The default is "SideQuest handles all incoming POs with the purchase-orders label." If you only want SideQuest on POs from specific customers, set up Gmail filters that auto-label only those customer domains. Or label manually per-PO.

Remove the purchase-orders label from the email. If you haven't asked Claude to draft the Estimate yet, the connector won't see the PO. If a draft already exists locally, ask Claude to discard it. If you've already submitted the Estimate to QuickBooks, open it in QB and either edit or delete it the way you would any other Estimate.

Three ways, all free, none require software. Sender filter for repeat customers, a Gmail plus-alias for forwarded stragglers, or a manual label click for one-offs. Copy-paste filter recipes live on the forwarding a stray PO page.

No. SideQuest runs inside the free Claude Desktop app — the same app you download from claude.ai/download. No Anthropic API key, no paid plan, no Enterprise subscription. The only Anthropic charge possible is if you opt in to handwriting OCR for scanned POs (about $0.20/month for typical volume).

SideQuest is built as a Model Context Protocol connector. Claude Desktop is the host that runs it. You ask Claude things like "list my incoming POs and draft estimates" and Claude calls SideQuest's tools, shows you the drafts in the chat window, and submits them to QuickBooks when you tell it to. The chat IS the review surface — no separate dashboard, no web page, no extra app to learn.

The conversation with Claude is how you interact with the product, so Claude Desktop is required. The matching, pricing logic, and QuickBooks writes are deterministic Python in the connector — Claude is the interface, not the brain doing the math. If a non-Claude path matters to you, email [email protected] — we're tracking interest in a standalone web review page.

Not today. SideQuest exposes itself via the Model Context Protocol, which Claude Desktop hosts natively. If ChatGPT or another agent adds MCP support later, the same connector should work there too.

Per company, flat monthly. Five tiers based on monthly PO volume: Free ($0, 20 POs), Starter ($79, 200 POs), Growth ($199, 1,000 POs), Scale ($499, 5,000 POs), Unlimited ($999, no cap). Same features across every tier — you pick the volume that fits. See the pricing page for the full comparison.

Soft handling: a warning email at 80% of your cap, a grace period at 100%, and a polite block at 125% asking you to upgrade. Nothing breaks mid-month. POs that come in after the block stay in your Gmail until you upgrade or until the next month starts.

Yes. Use the Stripe Customer Portal link in your most recent receipt email, or email [email protected]. You stay on your current tier until the end of the month you already paid for, then drop to the Free tier. No phone calls, no friction.