Docs menu

MCP Server

Give any MCP-compatible AI agent web search, claim verification and balance checks through a hosted remote MCP server — no local process to run. The API key travels in the URL or an auth header.

https://searchpipe.tech/mcp/?api_key=sp-YOUR_KEY

Transport: Streamable HTTP. Treat any URL containing your key as a credential.

Connect

Add the remote server URL to your client. Two authentication forms are supported — prefer whichever your client supports:

  • Key in the URL: https://searchpipe.tech/mcp/?api_key=sp-YOUR_KEY
  • Auth header: connect to https://searchpipe.tech/mcp/ with an Authorization: Bearer sp-YOUR_KEY header (or X-API-Key: sp-YOUR_KEY).
claude mcp add --transport http searchpipe "https://searchpipe.tech/mcp/?api_key=sp-YOUR_KEY"
Key safety. 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 API Keys if it leaks — a revoked key stops working at once. Agents can also self-configure from /agent-setup/SKILL.md.

Tools

Three tools are exposed. Tool names and parameters match the REST API; REST-only request metadata (request_id, response_time, usage, warnings, notices) remains null over MCP.

searchpipe_search

REST counterpart: POST /v1/search · 0.5 / 1 / 2 credits by search_depth

Execute a web search: multi-engine retrieval, full-text fetching, LLM reranking and an optional cited answer or structured output. Returns the same result list as the REST endpoint.

Accepts the same parameters as POST /v1/search except stream — see the Search API reference for every parameter and default. See the Search API reference.

searchpipe_verify

REST counterpart: POST /v1/verify · 1 credit per claim; failed calls are refunded

Verify a factual claim against live web evidence: four-label verdict (supported / partial / unsupported / inconclusive) with span-level quotes, machine-readable reason codes and source authority. Typically takes 5-15 seconds.

searchpipe_verify parameters
ParameterTypeDescription
claimstring · requiredThe claim (statement or citation) to verify; up to 2,000 characters.
modestring · default autoEvidence route: paper (DOI/arXiv), web, or auto — a DOI or arXiv signal picks the paper route.
evidence_urlsstring[] · optionalUp to 5 caller-supplied citation URLs to check.
max_evidenceinteger · default 5Maximum number of evidence items to return (1-10).

searchpipe_account_balance

REST counterpart: GET /v1/account/balance · Free to call — no credits charged, no usage logged

Check the remaining credit balance for the key: total balance, permanent credits, expiring subscription credits, upcoming credits and the next expiry date.

REST API vs MCP Server

Both paths run the same pipeline with the same credit balance — pick per consumer, not per project.

AspectREST APIMCP Server
Best forYour application or backendAn MCP-compatible agent or client
TransportHTTPS + JSON; optional SSEStreamable HTTP (MCP)
ParametersFull search request, including streamSearch parameters except stream
ResponseSearch content and request metadataSame search content; REST-only request_id, response_time, usage, warnings, and notices remain null
Billing0.5 / 1 / 2 credits by depthSame search costs and shared credit balance
CachingShared search cache; cache hits are chargedSame cache and billing behavior
Extraction is REST-only. POST /v1/extract and the verify batch endpoint are available through REST only — the hosted MCP server exposes the three tools above.

Client configuration

Step-by-step config snippets for Claude Code, Cursor, OpenAI Codex, OpenCode, Pi and custom MCP agents live in the agent setup guide; connection troubleshooting and key management live in the MCP setup guide. Programmatic integration paths (Python and TypeScript SDKs, directories) are cataloged in Integrations.