A memory for your own AI agent

Your AI does the looking. This is where the knowledge lives.

Snap the find, queue the question, and let your own AI agent do the research. Every item becomes a dossier — sources, price evidence, provenance — kept in a private library you can search and export.

Create your private library Sign in

Bring your own agent — Claude, ChatGPT, Cursor, or anything that speaks MCP. TPP never runs AI for you.

Capture in the field

At the yard sale, the antique booth, the estate line — snap a photo, jot a note, drop it in the queue.

Your agent researches

Connect your AI once over MCP + OAuth. Then just say “research my queue” — it identifies, prices, and sources each find.

The knowledge stays

Structured dossiers land in your private library: claims, corrections, value evidence — searchable, exportable, yours.

From evidence to a decision

See why the number changed.

TPP keeps the sold evidence, condition, fees, and uncertainty beside the recommendation. Your agent can update the conclusion without erasing the trail that produced it.

The example is synthetic product UI, not customer data or a claim about a real listing.

Synthetic example

Walnut record cabinet · 1960s

P50 · $185

Comparable evidence

Sold · local auction

Good · source linked

$165

Sold · marketplace

Very good · source linked

$195

Active listing

Asking · source linked

$240

Decision

Offer up to $140

$185 midpoint minus $25 transport and a visible uncertainty buffer. One high asking price is excluded from the sold range.

Range $165–$1953 evidence rows

For your AI chat

A starter prompt for your agent

Paste this into your AI chat so it checks for the TPP connector before assuming it can't help.

I have a ThePurchasePlanner account and I want you to help me use it. First, check whether ThePurchasePlanner or TPP tools are already available in this chat. If they are, use the MCP tools instead of scraping website pages. Read https://thepurchaseplanner.com/ai.txt and https://thepurchaseplanner.com/llms-full.txt if you need the current setup and tool contract. If the tools are connected, tell me what ThePurchasePlanner can help with and offer to research my queue. When researching my queue, list actionable captures, inspect photos with tpp_get_capture_images when there are several photos or tpp_get_capture_image for one photo, use prior research context for follow-ups, preserve any sessionId or ownershipStatus already on the capture, do the actual research, then prefer tpp_save_research_result to save the compact item profile, canonical item claims, any corrections/ruled-out facts, open questions, sourced dossier sections, source checks, run stats, reference images, session membership, and capture completion in one call. Always identify/classify the item as baseline work, even if identification is not selected as a Research Task. Use any selected Research Tasks on my captures as item-level research instructions to decide which extra sections to write. For follow-ups, treat the note/photos as additional context for the already researched item, then update the existing item and sections instead of making duplicates. If I selected "How much should I pay?", compare used prices by condition/venue and include new-price context when available. If I selected "Make a sales listing," draft a natural marketplace listing without AI-sounding wording and save it as a custom section with a stable section key. If you need to add any custom section, use a stable section key so it can be edited later. When updating old research, treat active item claims from tpp_get_research_context as the current source of truth. Dossier prose can be stale. Do not reintroduce a ruled-out fact unless you have stronger new evidence and explicitly log the correction/revival. When writing sections, include dependsOnClaimKeys for the canonical facts the section relies on so TPP can flag stale prose later. If a capture or item profile includes priceValues, use those explicit value evidence rows before trying to infer price from the note; treat older priceContext only as fallback data. If I tell you I paid for or own the item, set ownershipStatus to owned and add a priceValues row for the paid/observed amount when known. If I selected "Price guide by place," estimate Yard Sale, Facebook Marketplace, eBay, Antique store, and New exact or New comparable expectations; include a yard_sale row even when confidence is low; store the structured venue guide on the item profile and also write a pricing dossier section with sectionKey venue_price_guide. Do not pretend an exact new price exists when only comparable replacement pricing is available. If I ask to start or work inside a shopping session, use tpp_upsert_research_session or tpp_list_research_sessions. Preserve the sessionId for items in that antique trip, yard-sale run, estate-sale day, or multi-day hunt, and mark items owned/not owned when I tell you what I bought. If the tools are not connected, guide me to add the connector my AI app needs using https://thepurchaseplanner.com/ai and MCP URL https://thepurchaseplanner.com/mcp. Research the setup page and agent docs first so you do not guess. Setup uses browser OAuth. Do not ask me to paste a TPP API key, OAuth client ID, OAuth client secret, or password into chat. Explain that TPP exposes a set of tools (currently about two dozen), and my AI app may ask me to approve individual tool calls; I can approve them as they come up, or pre-approve/trust the TPP connector if my AI app offers that and I am comfortable. If my AI app has mobile limitations, tell me to use its desktop app or web version to add the connector, then I can usually use the connected tools from mobile afterward. After the connector is connected, ask me to enable it in this chat and then offer to research my queue or explain what TPP can help with.