You've got AI switched off.
That's a perfectly good way to use Ubikron. Switch it on to see what it can do.
An AI you can check up on.
Plenty of investigators don't trust AI, and fair enough. In Ubikron it's optional, it's recorded, and it never gets the last word.
Four ways to use it
It describes your clippings
Clip an image or a passage and it writes a description, so you can search by what's in it.
It knows the whole case
Ask about the investigation. Answers come from your pages, notes and graph, not the internet.
It browses alongside you
In Claude for Chrome it can browse, enrich and build the graph. Every page it visits and every call it makes is recorded.
It works from outside
Through MCP, agents like Claude Code, Codex or Gemini can drive all of Ubikron, the same way you do.
Anything you can do in Ubikron, an agent can do too.
Every function is exposed over MCP, so an agent can run an investigation the way you would. Whatever it does lands in the same record as your own work (but marked as AI).
Run projects
Create one, switch between them, read what a case is about and keep notes on it.
create-new-project switch-project understand-project save-project-memory
Browse
Open pages, scroll them and read their text and links, recorded like any page you visit.
go-to-link scroll-window get-page-text get-page-links
Search what was seen
Full-text search over every captured page, OCR text from screenshots, and tags.
search-history-full-text fetch-all-screenshot-ocr-text tag-entity
Build and edit the graph
Add, link, retype, recolor and delete nodes. What it adds stays grey, with dotted links, until you approve it.
add-graph-node link-graph-nodes edit-graph-node-type batch-update-graph
Enrich with Mimic
List the enrichments your server has, run them, and read the results.
list-mimic-services execute-mimic-service fetch-mimic-results
Write its report
Read yours, and write its own in a section kept apart from it.
get-user-report append-mcp-report edit-mcp-report
Talk to your graph
The answer comes from your own pages, notes and graph, not from the internet.
You don't have to trust it, because you can audit it.
It leaves the same trail you do: pages, timestamps, logged calls. Its report stays in its own section.
Connecting an agent over MCP
Point any MCP client at your server with an API key.
{
"mcpServers": {
"ubikron": {
"command": "npx",
"args": ["-y", "mcp-remote",
"https://<server>/api/mcp",
"--header", "X-Api-Key: <api-key>"]
}
}
}
Start here
ubikron-mcp.json: the MCP config above.
ubikron-skill.md: an OSINT investigator skill for agents. Methodology, documentation habits, when to stop and ask a human.
llms.txt: what Ubikron is, for the AI reading this page on your behalf.