A practical guide to writing a Swarms Marketplace listing that gets found and bought: titles, tags and categories, descriptions, sample outputs, the Verified badge and Security Score, how listings get surfaced, pricing, and keeping a listing fresh, with a pre-publish checklist.
The Swarms Marketplace lists more than 6,000 agents, prompts, tools, and MCP servers. A listing that is published but never found earns nothing, and the difference between found and buried usually comes down to a few fields on the launch page: the title, the tags, the first sentence of the description, and whether the page shows what the product actually does.
This guide covers each of those fields for every product type, grounded in how discovery works on the marketplace today. If you are publishing a prompt for the first time, pair it with Publish Your First Prompt in Ten Minutes, which covers the form itself.
Every decision below follows from the ways a listing can be found.
swarms.world/mcp lets other agents search by name, tag, category, language, required package, use case, or price.Write your name, tags, and description for a reader who is typing what they need, in their own words.
The name is the heaviest field in search and the only text a buyer is guaranteed to see on a card.
Lead with the job, then the domain. Put the noun a buyer would type first: "Clinical Trial Protocol Reviewer", "Shopify Order Reconciliation Agent". Exact and prefix matches rank first in the navigation search, so the first word matters most.
Keep it short enough for a card. Grid cards and table rows truncate long text. Three to seven words is a good range.
Avoid generic names. "AI Assistant" and "Research Agent" compete with hundreds of identical listings and tell a buyer nothing.
Do not start the name with "x402". Agents whose names begin with "x402" are pushed to the back of every registry listing by default, whatever the filters or sort order, and buyers can exclude them entirely. Version numbers, emoji, and all-caps spend characters a card does not have and match nothing anyone searches for.
| Weak title | Why it fails | Stronger title |
|---|---|---|
| AI Research Agent | Hundreds of identical names, no domain | Biotech Patent Prior-Art Researcher |
| My Awesome Prompt v2 | Says nothing about the job, version noise | Cold Email Sequence Writer for B2B SaaS |
| x402 Super Trading Bot 🚀 | x402 prefix sorts it to the back, emoji wastes space | Perpetuals Funding Rate Arbitrage Agent |
| Firecrawl Thing | Vague, not searchable by capability | Firecrawl MCP: Web Search, Crawl, and Scrape |
Tags are how the tag filter, the registry search, and the Autonomous Filter Search agent find you. Categories are how the industry filter and the MCP servers page find you. Set both.
Write tags the way buyers describe needs. A buyer types "customer support", "lead-gen", or "pdf-extraction", so your tags should be those phrases in kebab-case, roughly five to sixteen of them. The launch form's tag generator proposes a set from your name and description, refuses filler like ai, tool, helper, and useful, and mixes three kinds: industry (healthcare, fintech), type (research-agent, mcp-server), and capability (web-scraping, sql-generation). Start there, then add the terms you know your buyers use.
Match the catalog's vocabulary. A common tag like trading connects you to every trading search; a one-off like my-quant-system connects you to nothing. In the registry's table view every tag is clickable, so reuse the tags your strongest competitors carry, then add the ones that make you specific.
Pick the categories a buyer would filter by. The catalog spans about twenty, including Finance, Research, Developer, Security, Healthcare, Marketing, Sales, Customer Support, and MCP. If your product fits two, pick both. An agent published with an MCP server URL gets the MCP category automatically and appears on swarms.world/mcp-servers; a code-based MCP server published as a tool or prompt needs the MCP category selected by hand.
The description is the search text, the card text, and the sales page at once. Structure it in that order.
First sentence: the search snippet. Cards and table rows show a truncated description, and the registry search matches the full text. Make the first sentence a complete statement of what the product does and for whom, in the words a buyer would search: "Reviews vendor contracts for GDPR and SOC 2 clauses and returns a risk table with citations, for compliance teams at B2B software companies."
Second paragraph: inputs and outputs. For an agent, the input it expects (a URL, a CSV, a question) and the output format (a markdown report, JSON, a table). For a prompt, what the user pastes in and what the model returns. For a tool, the function in plain terms. For an MCP server, the marketplace reads your tools from the server and lists them for you (see Expanded MCP Support), so spend the words on what those tools accomplish together.
Third paragraph: requirements. The model it was built and tested on, any API keys or environment variables, package dependencies, and anything the agent touches (shell, network, a wallet). Buyers who run agents with their own keys look for this before they pay. The publishing judge that reviews every submission reads the full body for this kind of substance: a free listing needs a score of 2.5 out of 5 to publish, a paid one 3.5.
Then the use cases. The launch form's Use Cases section takes a title and a short description for each. Fill in two or three real ones; the marketplace MCP lets agents search listings by use case.
If you inspected an MCP server and the marketplace filled in the name and description from the server's metadata, rewrite that text in your own words. Keep the prose plain: the Summary tab on every listing condenses the page into two to four sentences, and a description that says concretely what the product does summarizes well.
Before paying, a buyer sees your image, description, ratings, upvotes, and purchase count. The body of a paid prompt or agent stays hidden until the purchase completes, so the strongest thing on the open part of the page is a sample of what the product produces.
Put the sample in a fenced code block, label it as an example, and mark anything abridged.
Set an image. Listings without one get a placeholder; listings with one get a 16:9 hero on the card and the page, and the launch form's image generator makes two banner options from your name and description.
Two trust signals on the marketplace are within your control.
Verified. Pro ($19.99 a month) and Premium members get a blue Verified label next to their @username on their profile at swarms.world/users/<your-username>. That is the page buyers land on when they click your name on a listing, where they decide whether to trust you with their money or their API keys. The badge means the account has an active paid plan; it is not an identity check and does not rate your products. See Get Verified on the Swarms Marketplace.
Security Score. Every prompt page has a Security tab and, once scanned, a Security Score: N tag in the banner next to tags like FRENZY and VAULTED. The score runs 0 to 100, higher is safer, with a green shield at 80 and above and a red shield below. It comes from SkillScanner, 49 deterministic rules plus an agent review. Details are in Every Prompt Now Has a Security Score.
Two facts change how you publish. First, the first complete scan is the one buyers see: editing the prompt afterward does not trigger a rescan, so scan before you publish. Second, most findings in honest prompts come from a few habits: "ignore previous instructions" or "do not tell the user", shortened or plain http:// links, API keys in the text, instructions hidden in HTML comments or base64, and shell or network access with no stated reason. Say what the prompt does openly, write links in full over HTTPS, keep secrets out, and explain any sensitive behavior in the prompt itself.
SkillScanner is open source and runs locally with scanner.scan_text() on your prompt file, so a green score is something you can earn before anyone sees the listing.
The marketplace does not publish a ranking formula, so work from the signals that are visible.
Filters are deterministic: a listing with the right tags, categories, and pricing flag appears in every filter combination that describes it.
Buyers filter by price and cards display it, so pricing is part of the copy.
The Seller Portal is where you maintain what you publish.
Read the reviews and comments feed. It collects every review and comment on all your listings in one place. A buyer question in a comment is a sale waiting on an answer, and you are emailed when someone comments or upvotes.
Edit from the product page. Click any listing in the Seller Portal table to open and edit it. Update the description when you change the model, add a dependency, or learn which phrasing buyers search for. Editing a prompt does not rescan it, so scan major revisions yourself. MCP listings refresh on their own: the marketplace re-inspects your server when the listing is viewed and the last check is over a day old.
Watch the 90-day earnings chart after every change. A new title, image, or tag set shows up in the bars within days.
Publish, open the registry, and click Refresh to confirm the listing is live. Then keep the Seller Portal open for the first comment.
A one-hour walkthrough for new visitors to swarms.world: find an agent with Autonomous Filter Search, read its listing, try it in Swarms Chat, pull it into your code through the Marketplace MCP and the Swarms API, and buy a paid listing by card, SOL, or USDC.
A hands-on tutorial: wrap the keyless Open-Meteo weather API in a Python MCP server with FastMCP over Streamable HTTP, test it, add a bearer-token check, ship it in Docker, and publish it on the Swarms Marketplace MCP Portal so agents can find and call it.
A migration guide for developers sitting on SKILL.md folders, .cursorrules files, or Custom GPT configurations: what maps to a prompt, agent, or tool listing on the Swarms Marketplace, what to strip first, how to import from private GitHub repositories, and what is worth charging for.