PersonIQ is a remote Model Context Protocol server that gives an AI assistant 18 tools for US people search, reverse phone lookup, property and owner records, and identity verification. One URL, no SDK, no local process. Pay per call in credits — never a subscription.
The PersonIQ MCP server is a hosted Model Context Protocol endpoint at https://mcp.personiq.io/mcp that lets an AI assistant run US people searches, reverse phone lookups, property and owner lookups, identity verification and source-cited web research directly inside a conversation. It is a production service rather than a demo, it works with Claude and ChatGPT as well as Cursor and any other MCP client, and it bills per call from a credit balance rather than by subscription. Because the same endpoint also speaks plain JSON-RPC over HTTP, a backend can call it without an MCP client at all.
PersonIQ is not a consumer reporting agency and this data is not a consumer report. It may not be used for employment, credit, insurance, housing or tenancy decisions. Read the acceptable use policy.
Every MCP client takes the same two things: the endpoint URL and your key. There is nothing to install, no package to add, and no process to keep running.
{
"mcpServers": {
"personiq": {
"url": "https://mcp.personiq.io/mcp",
"headers": {
"Authorization": "Bearer pk_live_YOUR_KEY"
}
}
}
}For chat connectors, PersonIQ is a full OAuth 2.1 authorization server with PKCE and dynamic client registration (RFC 7591). Point a connector at https://mcp.personiq.io/mcp, sign in, and approve scopes on a consent screen — no key ever gets pasted into a chat window.
You may untick any scope you do not want to grant. The server intersects what the consent screen returns with what the client asked for, so narrowing always works and widening is impossible.
Any instance serves any request, so there is no session to lose. Long jobs return a durable handle you poll — a dropped connection never costs you work you already paid for.
18 tools across six areas. Your assistant only sees the ones your credential is scoped for, so a narrow key produces a narrow tool list rather than a wall of permission errors.
Find a person by name, phone, email or street address and get the whole record back in one call — identity, phones, emails, address history, relatives and public financial signals. There is no separate unlock step.
Resolve any US number to the people behind it, or pull carrier, line type, location and spam risk without touching identity data at all.
Look up a single address, or filter a whole city, ZIP or county by equity, occupancy and distress. Then unlock the owner behind the parcel and skip trace the humans behind an LLC or trust.
Check whether a name, phone, email and address all describe ONE real person — cross-referenced against fraud and identity networks, not a background check. Returns a 0-100 trust score, a verdict, and a separate judgement on every value you send, so you learn WHICH detail belongs to someone else. Send up to 5 phones, 5 emails and 5 addresses: they cost the same as one, because the provider bills per call. A score of 0 is a verdict and is charged; only a provider failure refunds.
Source-cited research on a verified person from the public web — a fast pass, or an exhaustive background job you poll to completion.
Download a report as a file, poll long-running work, and read your own balance and scopes. Markdown export and the job and account reads are free; a PDF render costs one credit.
Full endpoint reference — parameters, return shapes and prices →
You do not call an endpoint. You describe what you want, and the assistant picks the tools, chains them, and tells you what it spent.
“Who is calling me from 661-555-0100?”
Reverse phone lookup, then carrier and spam risk on the same number.
“Find Jane Doe in Austin, TX and tell me which of her phone numbers is current.”
One person search, then reasoning over the returned phone history.
“Pull every absentee-owned single family home in ZIP 93312 with high equity, then unlock the top five owners.”
A filtered property search, then a quoted unlock on each parcel you pick.
“Does the phone number on this lead form actually belong to the name on it?”
An identity match with a confidence score, before you spend a dime on outreach.
“Research this person on the public web and cite your sources.”
A source-cited intelligence pass matched to the verified identity.
“Who owns 123 Main St, Bakersfield CA, and how do I reach them?”
An address lookup, an owner unlock, and a resident skip trace.
You still can — the same endpoint speaks JSON-RPC. But when the caller is a model rather than your code, the integration work mostly disappears.
Tool names, parameter schemas, prices and return shapes are published by the server. Nothing is documented in a PDF your integration has to keep up with.
Search, then unlock, then research, then verify — the assistant decides the order from what came back, rather than you writing that flow by hand.
Anything over one credit quotes first and only runs when the quote is confirmed. Every response reports its cost and your remaining balance.
A credential only sees the tools it holds scopes for. An agent cannot discover a capability it is not allowed to use, let alone call it.
Person search and reverse phone lookup refund automatically when they find nobody, so an exploratory agent does not bleed credits on dead ends. A negative identity verdict is an answer, so that one is charged.
Deep research, property unlocks and bulk traces return a job handle. Poll it whenever — the result is a durable record, not a live stream you can drop.
Ask for absentee-owned, high-equity or pre-foreclosure parcels in a ZIP, unlock the owners, and trace the humans behind the LLC — inside the assistant you already draft offers in.
Verify that the phone on a lead form belongs to the name on it before the number reaches a dialer, and enrich a record without leaving the workflow.
Cluster the public web into distinct candidate people, reveal the right one, and get a source-cited briefing rather than a page of raw records.
Ship a people-data feature without negotiating a data contract. One credential, published tool schemas, per-call pricing you can pass through.
You buy credits and spend them across every tool from one balance. Packs start at $50 for 1,000 credits, the cheapest billable call works out to about $0.02 at the best self-serve rate, and credits never expire.
4 of the 18 tools are free — reference data, and reads of things you already paid for.
An agent that can look anyone up is exactly the thing that needs hard limits, so ours are enforced in the server rather than written in a policy nobody reads.
PersonIQ is a remote Model Context Protocol (MCP) server for US people search, reverse phone lookup, property records and identity verification. You point an MCP-capable AI assistant at https://mcp.personiq.io/mcp, authenticate, and 18 tools appear inside the assistant. There is nothing to install and no local process to run.
Yes. It works with Claude Desktop, Claude Code and the Claude web app. Claude Desktop and Claude Code take the endpoint URL plus an API key header; the web app connects over OAuth as a custom connector, where you approve scopes on a consent screen instead of pasting a key.
Yes. PersonIQ implements OAuth 2.1 with PKCE and dynamic client registration (RFC 7591), which is what ChatGPT custom connectors require. You add the server URL, sign in, tick the scopes you want to grant, and the tools become available in the conversation.
No. Connecting a chat client is configuration only. If you are building a product rather than using an assistant, the same endpoint speaks JSON-RPC over HTTP, so any language that can POST can call it without an SDK.
There is no subscription and no minimum. You buy credits and spend them per call: packs start at $50 for 1,000 credits, and the cheapest billable call works out to about $0.02 at the top self-serve rate. 4 of the 18 tools are free. Credits never expire.
Not on the expensive calls. Every tool costing more than one credit returns a quote first and will not run until your agent passes back the confirmation token from that quote, so a multi-credit charge is always a deliberate second step. Cheap reads run directly, and every response reports what it cost and what your balance is now.
Person search and reverse phone lookup are automatically refunded when they find nobody — a miss costs us nothing at our data provider, so it costs you nothing. Identity verification is the deliberate exception: a score of 0 means the name and phone do not resolve to a known identity, which is a verdict you can act on and often the one you were buying, so it is charged and returned in full. Only a provider that never responds refunds there.
No. PersonIQ is not a consumer reporting agency and its data is not a consumer report. It may not be used for decisions about employment, credit, insurance, housing or tenancy under the FCRA, or for any DPPA-governed purpose. You declare a permissible purpose when you create a credential, and you are responsible for what your users and agents do with the data.
The data is the same; the integration is not. Over MCP the assistant reads the tool catalog itself, picks the right endpoint, fills the parameters, chains one call into the next and explains the result — so the glue code, the retry logic and the response parsing you would normally write are the model’s job instead of yours.
US public records and licensed commercial data: people, phones, addresses, relatives, property and parcel records, criminal and background records, plus AI research over the public web. Phone reference data such as area codes and carrier prefixes is free to call.
Create a key, paste one config block, and ask your first question. 4 tools are free to call while you are trying it out.