No. You keep the agent you already use — Claude Code, Codex, Cursor, OpenCode, Windsurf, GitHub Copilot, OpenClaw, Hermes or Claude on the web. You paste one line into it and it sets itself up: the file it fetches carries a section per kind of agent, and yours picks its own.
Reading the setup file, searching the catalog and inspecting an endpoint’s schema and price are all free and need no account. Only running an endpoint spends, and a free account starts with 50 credits and no card, so you can run real calls before deciding anything.
No — that is the point. The vendor access sits behind the skill, so there is no signup, no key to store and no separate invoice for each of the 32 services the catalog reaches. The vendor credentials are ours.
Every endpoint carries a real price and a pricing shape, and both show up in discovery. A dry run prices the exact call you are about to make and shows what would be sent, without spending anything.
A failed call is recorded at zero cost — you are not charged for a vendor’s bad day. Your agent gets the reason rather than a generic error, so it can widen the query or pick another endpoint. See Troubleshooting.
Yes, two ways: the most any single call may cost, and the most everything may cost in a day. Set both at agentmore.app/app/budget; a call over either is refused before anything is sent, so it costs nothing and a runaway loop ends at a number you chose. Detail: Spend caps.
Nothing you put in trains a model, and every account is scoped to itself — your runs, files and results are reachable only by you. Results come back to you intact; the only thing stripped from a vendor response is credentials.
Yes — that is the main way in. Paste the setup line into your own agent, connect the hosted MCP server, or call the HTTP API directly. All three reach the same endpoints against the same balance. The chat workspace is one consumer of them, not the product boundary.
Yes. Discovery and inspection are anonymous GET endpoints; running one is an authenticated POST with a key you create in the app. Full shape, response fields and statuses: HTTP API.
Usually it never fetched the file. Start a fresh session and paste the setup line on its own, then ask for a real outcome — “find the last 50 posts for this hashtag on TikTok”, not “use the skill”. If it still improvises, Set up your agent has the exact commands per agent, and you can run them yourself.
On the CLI path, no — a shell command has no browser handshake to hang a sign-in off, so it reads a key or it is not authenticated. Grab one at API keys and paste it when your agent asks. Agents connecting over MCP use the same key as a bearer token.
No. A free account can create a key and run endpoints against its 50 credits. A plan adds a balance that refreshes every cycle — 1,000 on Start, 3,000 on Pro, 5,000 on Master. Prices are at agentmore.app/pricing.
In the app, open the account menu, then Plans & pricing and Manage subscription. Cancelling stops the subscription and the monthly credits with it. See Plans & billing.