Lens

One tool call between an answer and its sources.

Lens MCP puts a source-checking tool inside the assistants your team already runs. The assistant calls it mid-answer and comes back with what held up, what did not and the passage behind every citation.

What comes back from one call.

Who disagrees
  • Disputing sources counted, not buried
  • Support and dispute on every answer
  • The split arrives in the reply
Citations that open
  • Every citation resolves to its passage
  • Verifying a claim is one click
  • No second research project
Sources, not echoes
  • Copies and syndication count once
  • Same-owner repeats fold into one voice
  • Over a billion passages behind it
Your stack unchanged
  • Your model and your prompts stay
  • Works with any MCP client
  • One entry in tool settings
Safe to switch on
  • Nothing you send trains a model
  • Per-query audit trail
  • Answers respect existing permissions

The check runs where the work already happens.

Nobody switches tools or opens a second tab. When a question turns on evidence, the assistant calls Lens, and the reply carries the sources behind it plus the ones that argue the other way.

In the assistant

Did the EU clear the merger and what conditions came with it?

Lens checked this answer

Cleared with conditions on the data business. Most sources describe the same remedy while a minority read its scope more narrowly. Both readings are quoted with the passage they came from.

212 distinct sources187 support25 dispute

Illustrative composition. The live tool runs on your question.

1
Search

Your question runs against the full index.

2
Collapse

Duplicates and syndication fold into one source each.

3
Count

Every source lands as support, dispute or silence.

4
Answer

Built from what held up, with the passage behind every claim.

5
Verify

The finished answer is checked claim by claim before the assistant replies.

What happens when the assistant calls.

Five steps run between the tool call and the reply. No answer skips the check.

Set it up once.

Create an account, generate a key and add the server to your assistant. Claude, Claude Code, IDE agents, internal copilots and anything else that speaks MCP can call it. Your model, your prompts and your workflow stay as they are.

Any MCP client
  • Claude, Claude Code and IDE agents
  • Internal copilots too
One key
  • Generate it in your account
  • One entry in tool settings
Built for teams
  • Point internal copilots at the same tool
  • Checked against your own documents

The next answer your team acts on should arrive with its sources.