An AI agent registry is a live inventory of every AI agent running in your company — each one mapped to its owner, model, tools, cost and last run. Fleece discovers your agents automatically and plots them onto one queryable graph, so the copilots, bots and side-project agents nobody tracked stop being shadow AI and start being something you can actually see.
You can't govern what you can't see.
Agents multiply faster than anyone is counting. A growth team wires up a scraper, support ships a triage bot, an engineer leaves a nightly agent running on a personal key. None of it is on a list anywhere.
By the time leadership asks how many AI agents the company runs, the honest answer is nobody knows. That sprawl is where budget leaks, stale agents linger, and a single forgotten key becomes a real risk.
- +Agents spun up across teams with no shared record
- +Shadow AI on personal keys, invisible to finance and security
- +Owners change roles and their agents keep running, orphaned
A live inventory, not a spreadsheet.
An AI agent registry is the catalog your stack never had: every agent, the model behind it, the tools it can call, the human accountable for it, and the last time it ran — captured as it changes, not typed into a doc that's stale by Friday.
Fleece builds it by watching what connects to the brain through MCP and folding each agent onto one live graph. New agent appears, it shows up; an agent goes quiet, you see that too.
Surface the shadow AI you didn't know existed.
Any client that speaks MCP — Claude Desktop, Cursor, Cline, Zed, the Fleece AI App or a custom agent — registers itself the moment it touches the brain. The agents teams built quietly stop hiding the day they query shared memory.
Each one lands on the graph next to the tools and people it touches, so discovery isn't a one-off audit you run and forget. The map stays current as your fleet grows.
Every agent has an owner and a price.
Click any agent node for live metrics: model, token spend, run-rate, owner and last run. The bot that hasn't fired in three weeks and the one quietly burning the most tokens both stop being invisible.
It's an inventory leadership can read at a glance — what's running, who's accountable, and where the spend goes — without chasing six teams for answers.
What the AI agent registry gives you.
Auto-discovery via MCP
Agents register themselves the moment they connect to the brain — no manual list to keep, no agent left off.
Live agent map
Every agent plotted next to the tools, integrations and people it touches, on one graph that updates as your fleet changes.
Shadow AI surfaced
Side-project agents and personal-key bots stop hiding the day they query shared memory. If it touches the brain, it's on the map.
Owner on every agent
Each node names the human accountable for it, so orphaned agents from role changes have somewhere to point.
Cost and run-rate per agent
Model, token spend, run-rate and last run for each agent — spot the stale ones and the budget-burners at a glance.
Queryable inventory
Ask any MCP client to list every agent on a model, owned by a team, or idle for weeks — the registry answers from the live graph.
Questions about the AI agent registry.
What is an AI agent registry?+
It's a live inventory of every AI agent running in your company, with each agent mapped to its owner, model, tools, cost and last run. Unlike a static spreadsheet, it updates as agents appear, change and go quiet, so the catalog reflects what's actually running right now.
How does Fleece discover our agents?+
Any client that speaks the Model Context Protocol registers itself the moment it connects to the brain — Claude Desktop, Cursor, Cline, Zed, the Fleece AI App or a custom agent. Each one is folded onto one live graph automatically, so you don't maintain the inventory by hand.
Can it surface shadow AI built by teams?+
Yes. The agents teams quietly spun up stop being invisible the day they query shared memory through MCP. Each appears on the graph beside the tools and people it touches, so discovery is continuous rather than a one-off audit.
What does it track per agent?+
Five things on every agent: the model behind it, token spend, run-rate, the owner accountable for it, and the last time it ran. Click any node to read them live and spot stale agents or quiet budget-burners.
Is this the same as governance and policy?+
No — the registry is about visibility: knowing every agent exists, who owns it and what it costs. Policy enforcement is a separate layer; you can't govern what you can't see, so the inventory comes first. The brain is local-first and cloud sync is end-to-end encrypted with AES-256-GCM.
Every agent. On one map.
Stop guessing how many AI agents you run. Map them once — owner, model, cost and all — and watch the registry keep itself current.