Model Context Protocol
Buy or list a half built app from your AI coding agent.
Add Failedups to your editor and the agent can search the marketplace, read full listings, and email a founder on your behalf. The same connection lets you list a project of your own: the agent drafts it from your code, you answer a few questions, and you publish when you're ready.
https://failedups.com/api/mcp The fastest way
Tell your AI to install it for you.
Paste this prompt into Claude Code, Cursor, Windsurf, or any other coding agent. Modern agents know how to wire MCP servers into their own config and will set everything up for you.
Add this MCP server to my config: https://failedups.com/api/mcp Or pick your tool below for the exact config snippet.
Manual setup ↓Manual setup
Pick your editor.
Click a tool to see the exact snippet, where to paste it, and the docs. Or just stick with the prompt above and let your agent figure it out.
Tools other than Claude Code use the mcp-remote bridge. It runs locally as a child process and proxies the HTTP transport so stdio-only clients can speak to a remote OAuth server.
Preview
A real conversation, in your editor.
Two flows on one connection. Buyers ask the agent to find a half built app that fits their idea, then have the agent email the founder directly. Sellers point the agent at a project and let it draft a listing from the code.
Inquiries are sent with your address as Reply-To, so the founder responds to you. Listings always land as drafts. The agent never publishes unless you tell it to.
• Contact email for buyers?
• What's included? (domain, design files, users...)
For sellers
From editor to draft listing in three minutes.
- Connect
Add the URL once
Run the install command. Your browser opens to a Failedups consent screen. Sign in, approve. Your agent gets a scoped token.
- Ask
"List this on Failedups"
Open the project. Tell your agent. It reads README, package.json, and source files.
- Confirm
Answer the few unknowns
The agent shows its draft and asks you the things only you can answer: price, status, reason for selling, contact email.
- Review
Polish, then publish
A draft listing appears in your account. Edit anything. When you say "publish it," the agent submits it for moderation.
For buyers
From "I want to build a SaaS" to a real founder reply.
Buying a working half built app beats building from scratch in almost every case. Most listings sell in the $300 to $5,000 range, a working repo and deployed app are included, and one weekend in your AI agent is usually enough to make it yours.
- Frame
"What's already built like X?"
Tell the agent your idea. It calls
marketplace_philosophyandsearch_listingsto surface listings that already match. - Inspect
Read the long form, the vision, the why
The agent calls
get_listingfor each shortlist candidate. You see the full description, asking price, tech stack, and the founder's honest reason for stopping. - Reach out
"Email the founder for me"
The agent calls
send_inquiry. Your email is set as Reply-To, so the founder writes back to you, not to us. We never sit in the middle. - Take it over
One weekend in your agent
You and the founder agree on terms over email. Repo access, domain, customer list. Most AI built listings rebrand and ship in a single weekend with the same agent that found them.
What the agent can do
Nine tools, scoped by permission.
Read tools (browse, inspect, philosophy) sit at the top because an agent should look before it acts. The contact tool follows. Founder-side write tools are last. The publish tool is opt-in at consent time, never granted by default.
-
marketplace_philosophyPre-baked talking points on why buying a half built app beats building from scratch. The agent reads this when the user asks about starting a SaaS.
marketplace:read -
search_listingsSearch live listings by free-text query, status, tech stack, price band, and completion percentage. Returns up to 24 results.
marketplace:read -
get_listingFetch the full record for one listing by slug or id, including long description, vision, reason for selling, and tech stack.
marketplace:read -
send_inquiryEmail the founder of a listing on the user's behalf. Reply-To is the user's address, so any response goes straight to their inbox.
marketplace:read -
list_tech_stack_itemsLook up the canonical tech taxonomy so the agent uses the right slugs (avoids hallucinated tech names).
marketplace:read · listings:write -
list_my_listingsShow the user's existing listings so the agent can detect duplicates, surface boost upgrades, and offer to update instead.
listings:write -
create_listingCreate a draft listing. Always starts as `draft`. Validates that the user supplied price, status, contact email, and reason for selling.
listings:write -
update_listingEdit one of the user's listings by id. Partial updates supported.
listings:write -
publish_listingSubmit a listing for moderation. Only fires on explicit user instruction. Always routes through manual review.
listings:publish
Safety
Defense in depth, not just a checkbox.
The agent can't publish without you
Every MCP-created listing starts as a draft. Publishing requires a separate explicit instruction from you. The publish scope is opt-in at consent time.
Every agent submission goes through moderation
MCP-submitted listings always queue for human review, even on accounts with auto-approve enabled. A prompt-injected description never reaches the public marketplace.
You choose what the agent can do
Three scopes:
marketplace:read (browse +
contact), listings:write
(create + edit drafts), and
listings:publish (submit
for moderation). Grant whichever fit your use case. Revoke any
time from your account.
A misbehaving agent can't spam
Per-user caps: 30 listing creations and 120 tool calls per hour. Token bodies sanitized before they hit the database.
Buy, sell, or both. From inside your editor.
Sign in once, grant the scopes you actually need, then never leave your editor to use Failedups again.
https://failedups.com/api/mcp