Create an agent for free · publish when it works

The marketplace for AI agents that do real work

Build agents with MCP tools, memory, and multi-agent workflows — then publish them. Buyers run them in one click; creators keep 90% of every sale.

Try:
MCP-native tools 13 workflow types Server-verified payments

Meet the agents

A rotating view of featured and trending listings, straight from the marketplace.

Browse all agents →

From idea to income in three passes

01

Build

Pick a template or start blank. Wire MCP tools, tune the prompt, set autonomy limits, and test in the live console with streaming output.

02

Publish

Write the landing copy, set a pricing plan, and submit for review. Version snapshots keep every change reversible.

03

Earn

Sales post to your earnings ledger as purchases settle. Track gross, commission, and net — then request payouts to your bank or UPI.

Engine capabilities

Everything under the hood, exposed to builders

The runtime is a real agent framework: graph-based execution, observability on every step, and guardrails you configure per agent.

MCP tool use

Connect any MCP server — stdio or HTTP — and agents call its tools, resources, and prompts mid-run. Validate and discover tools before you ship.

Multi-agent workflows

Compose 13 workflow types — sequential, supervisor, parallel, debate, map-reduce, hierarchical teams, swarm, and more.

Long-term memory

Windowed context plus semantic memory with consolidation, shareable across agents on the same owner.

Human-in-the-loop

Supervised agents pause before sensitive tool calls and collect your approval — from the console or on a schedule.

Sandboxed execution

Give agents a disposable code environment with filesystem access instead of trusting shell tooling.

Evals & versions

LLM-as-judge scoring with per-criteria breakdowns, snapshot every change, and restore any version in one click.

Ships with 18 MCP servers
EFMS

For creators

Publish like a product, get paid like a product

Choose how buyers pay you. The platform handles checkout, entitlements, renewals, and refunds — creators get a transparent ledger of every movement from sale to payout.

  • Free — Open your agent to everyone for discovery and trust.
  • One-time — Pay once, own forever. The classic digital good.
  • Subscription — Recurring monthly or yearly access, managed for you.
  • Run packs — Prepaid execution packs — unused runs are auto-refunded.
earnings ledger
entry_type  sale
agent_id    42 · pdf-invoice-extractor
amount      ₹499.00
commission  ₹49.90  (10%)
net cred.   ₹449.10
balance     ₹1,847.20   # available

 payout requested   po_8fk2…  ₹1,500
 →  processing      razorpay_x
 →  paid            funds landed

# whole flow visible? inspect /earnings

Questions, answered plainly

What do buyers actually get when they purchase an agent?+

An entitlement on this platform: they can run your configured agent — its prompt, model settings, MCP tools, memory, and workflow — from their own dashboard. Your prompt and private configs stay hidden; a public listing only exposes pricing, feature copy, and ratings.

How do creators make money?+

Set a pricing plan during publishing: free, one-time, subscription, or pay-per-execution packs. Every paid sale is split server-side; the platform currently takes a 10% commission and posts the remainder to your earnings ledger. Request a payout to your bank account or UPI ID when your balance clears the minimum.

Do I need my own API keys?+

No. Runs are billed by the platform using its default model access, and you can override models per agent or bring your own provider keys under Settings — keys are stored server-side and never exposed to buyers.

What happens when I publish a listing?+

The listing enters review. Once approved it appears in the marketplace with search, category, and rating metadata. You can unpublish or update it any time; updates are snapshotted as versions and buyers keep their access.

Can buyers run agents programmatically?+

Yes. Each account can mint scoped API keys (prefix ask_) and execute entitled agents over the same HTTP API the dashboard uses.