22 frequently asked questions about integration, billing, rate limits and result usage. Can't find your answer? Email support@searchpipe.tech or submit feedback from the console.
SearchPipe is a web search API and MCP server for developers building AI agents. A single call runs multi-engine retrieval, full-text fetching, LLM reranking and optional summarization, returning structured results with relevance scores that agents can consume directly.
SearchPipe returns cleaned page content, relevance scores and optional answers — not raw search-result HTML that you have to parse. Agents can consume the response directly, and failed requests are automatically refunded.
Two first-class paths, equally supported — pick whichever fits your stack: connect via MCP (any MCP-compatible agent or client — Claude Code, Cursor and more), or call the REST API POST /search directly from your custom agent in any language. Both run the same pipeline with the same search pipeline and credit billing. MCP excludes REST answer streaming and REST request metadata.
Credits: $0.005 = 1 credit for custom recharges (bulk tiers from $0.0040). Fast search costs 0.5 credits, basic 1 credit, and advanced 2 credits. The free plan includes 1,000 credits per UTC calendar month, renewed when a verified account first signs in that month. Recharged credits never expire; subscription credits are valid for 30 days.
Copy the MCP link from the dashboard into your client config — typically 1–3 minutes. Agents can also read /agent-setup/SKILL.md to configure themselves automatically.
Yes. SearchPipe provides a REST API (POST /search) that returns structured JSON results with relevance scores, extracted page content and optional AI-generated answers. Authenticate with an sp- prefixed API key.
Yes. SearchPipe runs a remote MCP server (streamable-http) at /mcp/. Any MCP-compatible agent or client can connect by adding the URL with an API key. Tool name: searchpipe_search.
SearchPipe is not a consumer search engine. It provides search and retrieval capabilities that developers can integrate into their own AI applications and agents.
Structured JSON with query, results (each containing url, title, content, score and optional raw_content), an optional answer with citation markers, and an ai_generated flag. The response shape is identical whether you call the REST API or the MCP server.
Every result carries an evidence object: a verbatim supporting quote with character offsets from the fetched page, a source-authority tier (official, press, academic, preprint, community or unknown) and fetch metadata (status code, redirects, final URL, fetch time). Advanced searches with include_answer enabled also return an answer audit: each statement in the generated answer is checked against its cited source with a four-label verdict, and the response carries an overall audit score.
inconclusive means the pipeline could not determine whether the evidence supports the claim — the candidate sources could not be fetched, sit behind a paywall, or did not contain enough checkable material. It is a first-class deliverable, not an error: an honest cannot-verify answer is billed like any other verdict, while a pipeline failure returns 503 and is refunded automatically.
Each call runs a real evidence pipeline: candidate retrieval, full-text fetching, sentence-boundary span alignment and a consistency check over the quotes. That is inherently slower than a plain web search, and a verification is capped at 30 seconds. Results are not cached, because a verdict depends on the live state of the cited pages.
A URL health check only reports transport facts about a link: status code, redirects and whether an archived snapshot exists. Verification answers a different question — does the cited source actually support the claim? Each evidence item carries url_health as one field, but the verdict comes from span-level quote matching and source authority, not from reachability.
The key lives in the client config file, so treat that file like a credential. Create a separate key for each agent client, and revoke it immediately in the dashboard if it leaks — a revoked key stops working at once.
A cold query runs the full pipeline — multi-engine retrieval, full-text fetching, LLM reranking — which usually takes a few seconds. Identical queries with identical parameters hit the cache (default TTL 300 seconds) and return in milliseconds.
Five by default; adjust with max_results (1–20). Results carry 0–1 relevance scores and are sorted from most to least relevant.
With include_answer enabled, the service generates a summary with citation markers based on the fetched page content, flagged with ai_generated=true.
No. When upstream retrieval fails (502) or output moderation rejects the content (400), the charge is automatically refunded (idempotently). Only successful requests are billed.
Yes: a sliding-window limiter (default 100 requests per minute, burst 20). Over-limit requests get a 429 with a Retry-After header. Contact us to raise the quota for high-volume use.
Checkout is handled by our merchant of record Creem, which accepts credit and debit cards, Apple Pay and Google Pay — the exact options shown depend on your country, currency and device. PayPal is not supported at the moment. You never enter card details on our site: payment is completed on the provider's hosted checkout page.
Credits from recharges never expire. Credits from monthly subscriptions (including renewals and upgrades) are valid for 30 days from the moment they land and expire automatically if unused.
Email support@searchpipe.tech with your account email, and include the order number if a payment is involved. If you can still sign in, the feedback form in the console lets you open a ticket with your order context attached automatically.
Email support@searchpipe.tech, sign in to submit a support ticket from the console, or let your agent read the setup guide and configure itself.