Blocket MCP server: let your AI agent search second-hand items in Sweden
Wield is an MCP server that gives your AI agent one Blocket tool, blocket.se/search-marketplace. The tool searches Blocket for second-hand items for sale, free giveaways or wanted ads. It returns title, price in SEK, place, when the ad was published, seller type, shipping and buy-now flags, map position and a link to each ad. It runs in your own browser, and no login is needed.
Blocket Search Blocket for second-hand items
Read-only, no loginPasses the daily check
Parameters (8)
What your agent gets
| Recipe | What it does | Login | Status (7 Nov 2026) |
|---|---|---|---|
blocket.se/search-marketplace | Searches for-sale ads, giveaways and wanted ads by query, price, condition, seller type and sort | No | OK, 6 of 6 checks, 1 request, 470 ms |
The tool takes query as its one required value, for example cykel (bicycle) or soffa (sofa). The other filters are optional: a price range in SEK, condition, seller (any, private or business), type (for_sale, giveaway or wanted), sort and page. Each run returns one results page of 53 ads, with total, last_page and the listings.
How a request flows
- You ask your agent: "Find a used bicycle under 2,000 SEK on Blocket, newest first."
- The agent calls
find_recipe, thenrunwithblocket.se/search-marketplaceandquery: "cykel",max_price: 2000andsort: "newest". wield-mcppasses the run through the local bridge to the Wield extension.- The extension sends the request from a blocket.se tab, so it runs the way your browser would run it.
- The ads go back to your agent. Nothing passes through wield.run.
In our check on 7 November 2026, a bicycle search under 2,000 SEK sorted by newest made 1 request and finished in 470 ms, passing 6 of 6 checks.
What leaves your computer
The only host the recipe calls is www.blocket.se. Each run makes one request, the same search the Blocket page makes. The recipe sends a sort order only when you ask for something other than relevance. Requests go out with the session your browser already has. Wield logs each request on your computer only, without bodies or query values. Telemetry is off by default. Wield never reads, copies or sends your cookies or passwords.
Blocket's terms, in plain words
We have not reviewed Blocket's terms for automated access, so we cannot say whether this use is allowed. What the recipe does: it makes one request per call, the same search the Blocket page makes. robots.txt allows the search and its API, but it disallows URLs with a sort parameter for bots. For that reason the recipe adds a sort order only when you ask for one. The recipe sends no messages, offers or favourites. This is not legal advice. Check Blocket's terms yourself before you automate anything else.
Limits and what is not working yet
- Read only. The tool searches. It does not send messages, make offers or save favourites.
- Relevance is the default sort. Other orders are sent only when you ask for them.
- 53 ads per page. Use
pageto read further pages.last_pageshows where the results end. - Condition has five values.
newmeans unopened,like_new,good,usedmeans visibly used, andneeds_repair. - Run limits apply. A run has at most 12 steps, 40 requests, 5 MB per response and 60 seconds.
FAQ
Is there an official Blocket MCP server?
The Blocket tool is a Wield recipe, made by Wield and not by Blocket. This page does not say whether Blocket offers an API or allows automated access. Check the site's terms.
Which AI apps work with it?
Any MCP client. npx -y wield-mcp init sets up Claude Desktop, Claude Code, Cursor and Codex. Other MCP clients work with the standard stdio config.
Does it need a Blocket login?
No. The search needs no account.
Can the agent message a seller or make an offer?
No. The recipe only reads search results. It sends no messages, offers or favourites.
How do I search free items?
Set type to giveaway. In our check, a giveaway search for "soffa" with private sellers returned 731 hits, and the listings showed a price of 0.
How does it compare with a browser agent?
Wield's preliminary benchmark, as of 7 November 2026, has no task for this recipe, so we have no speed or success figures to publish for it. The method and the tasks we did measure are on the benchmark page.