> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getunbound.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Discovery — Inventory
> Every AI tool, MCP server, config file and skill found across your devices
| | |
| ----------- | ------------------------------- |
| **URL** | `/ai-tools-discovery/inventory` |
| **Sidebar** | AI Tools Discovery → Inventory |
| **Roles** | Admin · Analytics Viewer |
Inventory is the catalogue of what discovery found on your developers' machines, sanctioned or not.
## What's on the page
### Sub-tabs
| Sub-tab | URL | Contents |
| ---------------- | -------------------------------------------- | --------------------------------- |
| **Tools** | `/ai-tools-discovery/inventory/tools` | Every AI coding tool found |
| **MCP Servers** | `/ai-tools-discovery/inventory/mcp-servers` | MCP servers configured on devices |
| **Skills** | `/ai-tools-discovery/inventory/skills` | Skills available to agents |
| **Config Files** | `/ai-tools-discovery/inventory/config-files` | Rules and instruction files |
Opening **Inventory** without a sub-tab lands on **Tools**.
### Tools
The tools found across your fleet run down the left, each with its plan and how many profiles carry it. Selecting one shows **Profiles Discovered** and **Config Files Scanned** for that tool, above the developers using it.
| Column | Contents |
| ------------------ | ------------------------------------------------ |
| **User** | The developer the installation belongs to |
| **Device Profile** | The account on the machine it was found under |
| **Device** | The machine itself |
| **Email** | Their address |
| **Config Files** | How many rules files came with that installation |
The **plan** shown against a tool is the tier the developer is actually signed in with on that device — not the seat you bought them. Where the two differ, that session is running under consumer terms your agreement does not cover. Reported for Claude Code, Cursor — the editor and the CLI — and the Auggie CLI; no other tool shows one. An **Account Access** policy is how you stop a session signed in with an unsanctioned account — see [Security Policies](/policies/security-policies).
### MCP Servers
**Publishers**, **Agents** and **Top Users** charts, above a table that folds the same server published under different names into one group — anything that does not group sits under **Ungrouped**. A group expands to the servers inside it.
| Column | Contents |
| ----------- | ------------------------------------------------------------- |
| **Group** | The group, or the server within it, with its publisher status |
| **Used In** | The tools the server is configured in |
| **Uses** | How many developers have it |
### Skills and Config Files
Both are one row per distinct file, with the same columns.
| Column | Contents |
| ------------ | ----------------------------- |
| **Name** | The file |
| **Category** | What it was classified as |
| **Users** | The developers carrying it |
| **Tool** | The AI tool that reads it |
| **Risk** | Its risk level, highest first |
Config Files covers the rules and instruction files the tools on a device read, whatever each one is called — global files and per-project files alike, including the ones a tool keeps somewhere other than its best-known location.
## Filters and controls
| Filter | Values | Where |
| ------------------------------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------- |
| **Status** | Shadow · Sanctioned | Tools |
| **Users**, **MCP Servers**, **Group**, **Used In**, **Publisher Status** | The developers, servers, groups, tools and publisher standing in your fleet | MCP Servers |
| **Category** | What the file is about: Security best practise · Code quality and style · Performance & Efficiency · Reliability · Testing & Error Handling · Maintainability · Framework or Tech Stack Conventions · Version Control & CI/CD · Code Generation & Automation · Documentation & Knowledge · Development Workflow. A file can carry several | Skills · Config Files |
| **Users** | Specific developers | Skills · Config Files |
| **Tool** | The AI tools in your fleet | Skills · Config Files |
| **Risk Level** | Critical · High · Medium · Low | Skills · Config Files |
The Tools sub-tab has a tool picker at the head of its list; MCP Servers, Skills and Config Files each have a search box.
## What you can do
| Action | Result |
| ---------------------------- | ----------------------------------------------------------------------------------------------------------------------- |
| Select a developer on a tool | Opens that installation in a side panel: **Overview**, then **MCP Servers**, **Tool Rules**, **Skills** and **Plugins** |
| Select a config file | Opens its own page at `/ai-tools-discovery/inventory/config-files/[contentHash]` |
| Select a skill | Opens its own page at `/ai-tools-discovery/inventory/skills/[contentHash]` |
### File page
| Section | Contents |
| ------------------------------------------ | --------------------------------------------------------------------------------------- |
| **Risk Assessment** | The risk level and the findings behind it, each of which can be marked a false positive |
| **Scope** and **Tool** | Whether it is a global or project file, and the tool that reads it |
| **Categories** | What it was classified as |
| **Used by** | Every developer carrying it |
| **Configuration** | How it is set up |
| **Skill Content** or **Tool Rule Content** | The file itself |
**Risk Assessment** names what was found, not just how bad it is. Findings fall into a fixed set of types: prompt injection, data exfiltration, credential access, code execution, hardcoded secrets, obfuscation, agent directives that override the developer's own instructions, self-modification, social engineering, supply-chain references, and text hidden with invisible Unicode. A file can carry findings of more than one type, and its level follows the most severe. Marking a finding a false positive removes it from that file's score everywhere the file appears.
**Categories** is a separate axis — what the file is *about*, not what is wrong with it — so a file can be categorised Testing & Error Handling and still carry a Critical finding.
Config file and skill addresses identify the **content**, not the device. Two machines carrying an identical rules file open the same page, so that page shows how widely a given instruction has spread across your fleet.
## Messages you may see
| Message | What it means |
| ---------------------------------- | --------------------------------------------------------------- |
| `Failed to load config files` | The Config Files list could not be fetched. Refresh to retry |
| `Failed to load skills` | The Skills list could not be fetched |
| `Failed to load tool rule details` | The list loaded but that config file's page could not be opened |
| `Failed to load skill details` | The list loaded but that skill's page could not be opened |
## Related
| Page | For |
| --------------------------------------------------- | --------------------------------------------- |
| [Discovery — Summary](/dashboard/discovery/summary) | The fleet overview |
| [MCP Connections](/dashboard/mcp-connections) | MCP clients you connected to Unbound yourself |