Help: every question, answered
Every question about Arettic in one place: what it is, giving AI agents APIs, MCP, choosing providers, spending limits, pricing, refunds and data.
- Arettic basics (14)
- Giving AI agents APIs (11)
- MCP and connecting (4)
- Choosing a provider (7)
- Spending limits and safety (7)
- Pricing and credits (12)
- Failures, refunds and disputes (3)
- Data and privacy (3)
Arettic basics
What is Arettic?
One API key for every tool your AI agent needs: web search, page reads, contacts, company data and more, from many providers on one balance. Failed calls are free.
Will I be charged if a call fails?
No. A call the provider fails costs nothing. For guaranteed jobs, a result that misses the published check, such as a catch-all email or a blocked page, is refunded too.
Do I need an account with each provider?
No. Your Arettic key covers every provider, with one balance and one bill.
Can I use Arettic with Claude Code or Cursor?
Yes. Connect any MCP client to https://api.arettic.com/mcp with your agent key, or call the REST API.
How is this different from picking a tool myself?
Arettic ranks tools by public scores for your exact task, runs the best one, and refunds results that fail the check.
Can my agent overspend?
No. Each key has a monthly limit ($50 by default) and asks before big purchases ($20 by default). Both are enforced on our side.
Can a provider pay to rank higher?
No. Tools are ranked by score, then price. Paying never changes a score, rank or routing decision.
Where can I try it before I build?
In the playground. It runs examples without a key.
Which jobs come with a guarantee?
6 jobs: verified email, verify email, company profile, person profile, web search and read a page. Each has a published pass rule, and a result that misses it is refunded.
What is a pass rule?
A published check a result must meet before it's charged. A found email, for example, passes only if it verifies as valid, not catch-all or unknown. Every rule is on Pass rules.
What is a credit?
The unit prices are shown in. 1,000 credits = $1, bought in US dollars with sales tax included.
Is there a free trial?
Yes. New accounts get $1 of free credit with no card, and $10 more after a demo call. Trial credits are spent first and last 30 days.
Can I test without spending credits?
Yes. Test keys (sk_test_…) call mock providers that give the same answer every time and never spend credits, so you can build both the pass and the fail path.
How are scores calculated?
Each tool gets a 0 to 100 score per job and region: 40% accuracy, 25% pass rate, 20% audit precision, 10% reliability and 5% speed, minus a penalty for upheld disputes. Every tool runs the same test cases each week. The formula and every input are on How scores work.
Giving AI agents APIs
What is an API marketplace for AI agents?
A place where an agent can find, pay for and call outside APIs through one account, instead of a developer signing up with each provider. Arettic is one: one key, one balance, public scores, and no charge for calls that fail.
Can AI agents use APIs without custom integrations?
Yes, through a gateway or an MCP server that already speaks each provider's API. With Arettic, your agent sends the provider's own parameters to one endpoint, POST /v1/call, with one key.
What is the difference between MCP and an API?
An API is how software calls a service. MCP (Model Context Protocol) is a standard way for an AI agent to find and call tools, and the tools often wrap APIs. Arettic offers both: a REST API and an MCP server with the same tools.
How do AI agents discover tools?
From a list they can search while they run: an MCP server's tool list, an API catalogue or a recommendation service. On Arettic, search_endpoints searches the catalogue and recommend ranks the tools for a job.
Can AI agents buy APIs on their own?
Yes, within limits a person sets. An Arettic agent key can buy results and call endpoints, but only inside its monthly budget, and anything over its approval threshold waits for a person.
How do you control API costs across many agents?
Give each agent its own key with its own budget and approval threshold, and pay only for calls that work. Agents are unlimited on every Arettic plan.
How do I manage API keys for autonomous agents?
Keep provider keys away from the agent. With Arettic the agent holds one revocable agent key, and provider keys stay on our side and are never sent to it. Add an IP allowlist, and revoke the key at once if it leaks.
What does an AI agent need to use outside services?
Four things: a way to find the right service, a key it can use safely, a budget it can't exceed, and a check that tells a good result from a bad one. Arettic gives all four through one key.
Do I need a separate API key for each provider?
Not with a gateway. On Arettic, one agent key covers every provider in the catalogue.
Should I use MCP or REST?
Use MCP when your agent runs in an MCP client like Claude Code or Cursor. Use REST from your own code. Both reach the same tools.
What happens when a provider is down?
On Arettic, a call the provider fails costs nothing, and guaranteed jobs can fall back to the next-ranked tool.
MCP and connecting
Is the Arettic MCP server free?
Connecting is free, and so are recommendations and the catalogue. You pay for results that pass and for endpoint calls that work. New accounts get $1 of free credit.
Which MCP clients does it work with?
Claude Code, Cursor and any client that supports Streamable HTTP with a custom header. Clients that only run local servers, like Claude Desktop, can use the bridge once @arettic/mcp is on npm.
Can my agent overspend through MCP?
No. Each agent key has a monthly budget ($50 by default) and an approval threshold ($20 by default). Over either, the purchase waits for a person.
How many requests can it make?
600 requests a minute per key to /mcp, within your plan's shared request limit.
Choosing a provider
What is API discovery for AI agents?
It lets an agent find the right API for a job at runtime, instead of a developer hardcoding one provider. On Arettic, recommend returns the tools for a job ranked by score, then price.
Is recommend free?
Yes. It uses no credits. It counts toward your plan's daily lookup quota.
Can my agent use a provider that isn't the top pick?
Yes. The ranking is advice. Your agent can send any purchasable tool_id to execute.
Does Arettic rank the tools it sells higher?
No. Whether a tool can be bought here is a field in the answer, not a factor in the rank.
Does it work over MCP?
Yes. The recommend tool on the MCP server does the same.
Is choosing at runtime slower?
It adds one call. recommend is free, and your agent can reuse its answer for the same job and region.
Can I force a provider?
Yes. Send any purchasable tool_id to execute. On Pro and up, owners can pin a tool so it's tried first.
Spending limits and safety
How do I stop an AI agent from overspending?
Give it a key whose budget and approval threshold are enforced by the API, not by the agent's own code. On Arettic both are on by default.
Can the agent raise its own limits?
No. An agent key can read its limits but never change them. Only an owner, or an org key with write access, can.
What happens when an agent hits its budget?
The purchase waits for an owner to approve it. Nothing is charged until they do.
Do endpoint calls count toward the budget?
Yes. Endpoint calls and guaranteed jobs share the same monthly budget.
Can I limit where a key is used from?
Yes. Add IP addresses or CIDR ranges to the key's allowlist.
What's a sensible budget for a new agent?
Start low and raise it once you've seen a month of spend. Arettic's default is $50 a month and $20 a request, and you can change both at any time.
Do failed calls use up the budget?
No. A failed call is refunded, so it doesn't count as spend.
Pricing and credits
What counts as a failed result?
Anything that doesn't pass the published check for its task type: a catch-all email, a record missing required fields, a blocked page. The hold is released and your agent gets the reason, not the data.
What can my agent buy today?
Six task types: find_verified_email, verify_email, enrich_company, enrich_person, web_search and extract_url. Each has one published check. New task types go live only when their result can be checked automatically, and their prices are on /prices.
Is asking which tool to use free?
Yes. Recommend and score lookups are free on every plan, and they cover tools we don't sell. You pay when your agent buys a result and it passes.
Do I miss anything on Pay as you go?
No. Pay as you go has the whole product: every tool and endpoint, unlimited agents and seats, segment scores, routing rules, private benchmarks, budgets, approvals, receipts and refunds. Pro and Scale only lower the price of each result, include some benchmark calls and answer faster.
When is a subscription worth it?
When you use more credits than the plan costs. Pro is $149 for 180,000 credits a month, so it pays for itself once your agents use about 150,000 credits a month (about 120,000 billed yearly). Scale is $499 for 600,000 and beats Pro from about 570,000 a month. Below that, Pay as you go is the cheapest way to use Arettic.
Is there a subscription?
Not on Pay as you go: you buy credit packs and spend them on passes. Pro and Scale are monthly or yearly plans that come with credits: unused monthly credits roll over one month. Yearly is two months free ($1,490 for Pro, $4,990 for Scale) and gets all twelve months of credits up front, valid for 12 months, with no rollover. You can move between plans at any time, prorated, and cancel any time, keeping the plan until the paid period ends. Max is invoiced.
How many agents can I run?
As many as you like, on every plan, with as many teammates as you like. Each agent has its own key, budget and approval threshold. All of an org's agents together share a per-minute request limit: 1,200 on Pay as you go, 3,000 on Pro, 6,000 on Scale.
Why does a result cost more than calling the provider myself?
Because you pay only for results that pass, and the work around the call is ours: picking the best-scoring tool, checking the answer, retrying and falling back, and the receipt. If you already know which provider you want and can check its results yourself, calling it directly is cheaper. Arettic is for agents that need a right answer without a person choosing the tool.
What if the provider is down?
We retry once on a server error or timeout. If it still fails, you're not charged and get provider_error. With fallback: true, we try the next-ranked tool within your max price instead.
Do you rank the tools you sell higher?
No. Tools are ranked by score, then price. Whether a tool can be bought through Arettic never changes its rank, and the scoring formula is public.
Do prices include tax?
Yes. Every price on this page includes sales tax, so you pay the figure shown. Your receipt shows the tax part.
Can I withdraw unused credits?
No. Credits are prepaid service for our catalog. They can't be withdrawn or moved to another account, except where the law requires a refund.
Failures, refunds and disputes
Should an agent retry a failed API call?
Once, after an error or a timeout. After that, switch provider or stop. Never retry an error that says the input is invalid.
Do I pay for failed calls on Arettic?
No. A call the provider fails costs nothing, and a guaranteed job that fails its check is refunded.
What if a charged result turns out to be wrong?
Open a dispute within 7 days with the receipt and the item. We decide within 48 hours against the stored result; if we don't, it's upheld. An upheld dispute is refunded and counts against the tool's score.
Data and privacy
How long do you keep my data?
Each purchase's input and result are stored encrypted with your org's own key for 7 days, so scores stay current and disputes can be checked, then deleted. After that we keep only hashes and non-personal signals.
Do you use my data to train models?
No. Nothing is shared between customers, cached for others or used to train models.
Can I ask you to delete my data?
Yes. Ask us and it goes at once.