Limits
Rate limits per key and per IP address, request and response sizes, tool timeouts, spend caps, and the few rules on inputs.
Rate limits
| Limit | Value | When it is hit |
|---|---|---|
| Paid calls per key | 300 per minute (a setting; fixed one-minute windows shared by every API instance) | rate_limited, HTTP 429, Retry-After, ₹0. Applies to POST /v1/run, run_tool and the headline MCP tools. X-RateLimit-Limit and X-RateLimit-Remaining are on every POST /v1/run response. Free calls (quote, discover, tools, balance, runs) do not count. |
| Failed authentication per IP address | 20 per minute | rate_limited, HTTP 429, Retry-After. Counts invalid keys and refused sign-in tokens. |
| New accounts per IP address | 10 per day (India time) | The dashboard says so and asks you to try again tomorrow. Only an account that is really created uses up a place. |
| Paid calls running at once per account | 10 (a setting) | too_many_in_flight, HTTP 429, Retry-After, ₹0: refused before anything is held, and no call is recorded. Wait for a running call to finish, then retry with the same idempotency key. The source's own limits are relayed as upstream_busy or upstream_unavailable (₹0, retryable); verify_email's source allows about 10 checks a second. |
| Live API keys per account | 20 (a setting) | key_limit_reached, HTTP 409, when making one more. Revoke a key you no longer use first. |
| Pages that need no key, per IP address | /health 120 a minute, /.well-known/ 3000, pictures under /i/ 1200 a minute, unknown /go/ links 30 a minute | rate_limited, HTTP 429, Retry-After. |
Sizes
| Limit | Value |
|---|---|
| Request body | 1 MB; over it, body_too_large (HTTP 413). Tool inputs are a few hundred bytes. |
tool | 1 to 64 characters |
idempotency_key | 1 to 128 characters |
client_tag | 1 to 64 characters |
| Response over REST | unlimited: data is the source's whole answer, however large (an Instagram page runs to hundreds of KB) |
| Response over MCP | 48,000 characters of JSON. Over that, whole items are dropped from the end of the largest list, then the longest strings shortened, so the result is still valid JSON, with a _truncated note (at, kept_items, total_items, shortened_strings). Billing settled on the full body before the cut. Ask for fewer results, or use REST for everything. |
Timeouts
Each tool has its own timeout, shown as timeout_seconds by GET /v1/tools/:id and on its tool page. After it the call is upstream_timeout: ₹0, retryable with the same idempotency key. Today's range is 30 seconds for most tools to 180 seconds for check_ai_visibility.
Money limits
| Limit | Value |
|---|---|
| Spend caps | A daily and a monthly cap per key, and a daily and a monthly cap per team, set in the dashboard in paise. Checked with the hold, in the same transaction, with the team row locked. A refused call is budget_exceeded (429) with cap: key_daily, key_monthly, team_daily, team_monthly. Days and months are India time. |
| Confirmation threshold | ₹200 (a setting). Above it a call needs confirm: true inside input; without it, confirmation_required (428), ₹0. |
| Hold | The estimate plus 50 percent, never less than ₹0.38. On settlement the real charge is taken and the unused part released. See if a source bills more than expected. A hold left open expires after one hour (24 hours for a slow job) and is swept back to the balance. |
| Smallest margin | ₹0.10 a call above the source's cost; there is no flat minimum charge |
| Automatic refunds | Up to ₹200 per team per calendar month through Report bad result, within 7 days of the call; more goes to review. |
| Welcome credit | ₹100, once per person and email inbox, no card. Addresses that reach one inbox count as one: capital letters and anything after a + never make a new address, and neither do dots in a Gmail address (jane.example@gmail.com and janeexample+2@googlemail.com are one). Sign-up refuses a second account for an inbox that already has one, and temporary (throwaway) email addresses cannot sign up. |
Rules on inputs
- One query, one handle, one address, one URL, one subreddit per call. No tool takes a list.
- Per-result tools hold for the number of results you ask for, rounded up to the source's billing block, never more than the ten-result worst case; without a count the worst case is held. Of today's 268 tools
domain_searchis priced per result (per started block of ten) and a few others per item returned (get_ad_historyper ad, for one; each tool's page says so); the rest hold their flat price plus the buffer. confirminsideinputis stripped before validation and never reaches the source.- Results are for your own, lawful work: no bulk unsolicited messaging, no reselling raw data, no lookups of minors. Acceptable Use Policy.
This page as Markdown: /docs/limits.md. Something wrong or missing: hello@parel.si.