Documentation

Know whether an AI answer holds up before you rely on it

FALCON Verify checks each claim in an AI's answer against the evidence it was given and tells you what holds up. It works inside Claude and ChatGPT, and in your own software through one address. Nothing to install.

Start in three steps 1. Install FALCON Verify in Claude or ChatGPT. 2. Sign in once. 3. After any answer, say “FALCON, verify that.” https://falcon.1-i.ai/mcp The same address for your own software and other MCP clients.

01Connect an assistant

FALCON Verify is available directly in Claude and ChatGPT. Install it from their directories, sign in to your FALCON Verify account once, and verify answers without leaving the conversation.

Connector

Claude

FALCON Verify is available directly in Claude's connector directory, on claude.ai and in Claude Desktop.

Open FALCON Verify in Claude →

Get started

In Claude, go to Customize → Connectors → Search, then search for FALCON Verify.
Select FALCON Verify and click Connect to Claude.
Claude opens the FALCON Verify sign-in page. Sign in to your FALCON Verify account and grant access.
Return to Claude and continue your conversation.

Then just say“FALCON, verify that.”

Plugin

ChatGPT

FALCON Verify is available directly from the ChatGPT plugin directory.

Open FALCON Verify in ChatGPT →

Get started

In ChatGPT, open Plugins and search for FALCON Verify.
Open FALCON Verify and click Install plugin.
ChatGPT opens the FALCON Verify sign-in page. Sign in to your FALCON Verify account and grant access.
Return to ChatGPT and continue your conversation.

Then just say“FALCON, verify that.”

Developers

Your software, and other MCP clients

Bring the same verification into your applications, agents and workflows. Sign in at https://falcon.1-i.ai/verify/ and create a token on the Connect page; every address below takes it as Authorization: Bearer.

REST APIPOST https://falcon.1-i.ai/v1/falcon/verify

MCPhttps://falcon.1-i.ai/mcp

Other clients: Gemini CLI, Cursor, any MCP client, your own code

Every other MCP client uses the address https://falcon.1-i.ai/mcp with a connector token created on the Connect page; export it as FALCON_TOKEN and refer to it as $FALCON_TOKEN rather than pasting it into a file you commit.
Gemini CLI: in ~/.gemini/settings.json add falcon under mcpServers with "httpUrl": "https://falcon.1-i.ai/mcp" and "headers": {"Authorization": "Bearer $FALCON_TOKEN"}}; restart gemini and /mcp lists falcon as connected.
Cursor: in ~/.cursor/mcp.json add falcon under mcpServers with "url": "https://falcon.1-i.ai/mcp" and "headers": {"Authorization": "Bearer $FALCON_TOKEN"}}; restart Cursor and switch falcon on in Settings → MCP.
Any mcp.json client: type http, the address above, and the same Authorization header.
Your own code: POST https://falcon.1-i.ai/v1/falcon/verify with the header Authorization: Bearer $FALCON_TOKEN and a JSON body of output and evidence; the Python SDK (pip install falcon-os) wraps it as FalconClient.verify.

02Confirm the connection

Ask the assistant: What FALCON tools do you have, and is FALCON ready? It should list falcon_verify and say FALCON is ready.
Ask it to verify a short answer. Every FALCON result ends with a line that begins with FALCON's mark and reads like Verified by FALCON Verify · run vf_… · 21 of 21 material claims decided, 0 unsupported (0%) · protocol v1.28 · chain of 11 entries verified. Verdicts without that line did not come from FALCON.
The Connect page in your workspace shows the most recent call once one has arrived.

03What to say

Verify the last answer
falcon verify the above response
Check a draft against a document
Use FALCON Verify to check this summary against the attached document. Send the document text as the evidence. List every unsupported claim and what would settle it.
Verify with your own claim list (free on every plan)
Run falcon_verify with claims = the statements I list below, in the output's own words, so no model extraction is needed: …
Confirm the connection
What FALCON tools do you have, and is FALCON ready?

04The tools the assistant gains

`falcon_verify`	the check: an answer, the evidence, a verdict per claim, the attribution
`falcon_verify_status`	a queued verification's state and final rate
`falcon_tool_readiness`	whether FALCON can run right now: model, store, plan
`falcon_whoami`	which tenant and credential the assistant is acting as
`falcon_account`	the plan, the allowance left this month, when it resets

The reading tools are marked read-only, so an assistant may call them without asking you each time. falcon_verify writes only its own record of the check and changes nothing else.

05The verdict

Know what holds up, and what doesn't. FALCON does not give you another confidence score. It checks each claim against the available evidence and tells you what the evidence supports.

✓SupportedSUPPORTEDThe available evidence supports the claim.

—Not supportedNOT SUPPORTEDThe claim is not supported by the available evidence.

×ContradictedCONTRADICTEDThe evidence says something different.

?InsufficientINSUFFICIENT EVIDENCEThe available evidence cannot confirm the claim, so FALCON does not guess.

A verification answers "is this supported by the evidence given", never "is this true". How checks are counted: one verification is up to 2,500 tokens of model work, counted at the model's own prices; a passage the model reads from its cache counts a tenth, so checking many statements against one document costs little more than checking one. A check whose statements you list yourself calls no model and is never counted.

06See it work

Three short demonstrations, each a person and the mistake caught before it cost anything.

Meet DiegoHow to catch AI hallucinations before pitching executivesA pitch deck's figures, checked against the sources before the room sees them.Watch on YouTube →Meet Maya, customer support agentHow to catch AI errors before they cost youA support answer verified against the policy it quotes, before it reaches the customer.Watch on YouTube →Meet Priya, finance analystHow to stop AI errors before executive meetingsA summary of the quarter, every number bound to the statement it came from.Watch on YouTube →

07Plans

Free — 25 verifications to start, once, with FALCON finding the statements; listing the statements yourself is free and unlimited. No card.
Starter — $8.99 a month, 150 verifications, then 9¢ each.
Personal — $19 a month, 400 verifications, then 8¢ each.
Team — $149 a month, 3,000 verifications pooled across 10 people, then 6¢ each.
Pay as you go — no monthly fee, 10¢ per verification.
Enterprise — a committed volume, single sign-on, a copy inside your own network, the offboarding certificate. Contact sales.

08Customer support

Write to us
contact@1-i.ai

Say which assistant you connected, what you asked, and the run id from the result's closing line if you have one. A person answers.

Before you write

No tools listed: start a new conversation after connecting.
Refused for allowance: the Usage page shows the month; listing the statements yourself is never counted.
A verdict you disagree with: open the run from the Connect page and read which evidence each statement was bound to.