Keep your AI agent skills organized in your personal cloud vault. Access them seamlessly across four surfaces: authenticated vault URLs, JSON REST API, MCP server, or the CLI tool.
- Authenticated vault URL: paste into any AI that fetches the web
- API & MCP server: for ChatGPT, Claude.ai, Gemini, and custom apps
- CLI tool: for Claude Code, Cursor, Windsurf, Antigravity, Codex, and more
- Four surfaces, one source of truth, one URL that stays the same everywhere
What SkillVault is
SkillVault is a personal cloud vault for AI agent skills. You store a skill once: a name, a description, and a body of instructions. SkillVault exposes your library across four surfaces simultaneously: an authenticated URL your agents can fetch, a JSON REST API your applications can call, an MCP server endpoint for native Model Context Protocol integrations, and a CLI tool that installs wrapper skills into local AI harnesses like Claude Code, Cursor, Windsurf, Aider, Antigravity, Codex, and DeepSeek Harness. One source of truth. Four delivery surfaces. The same skill library everywhere.
A skill vault is your secure, private, unified space for AI instructions. Your skills live in your personal vault, served securely under tokens you control and can rotate at any time. The primary vault URL you share stays consistent, while rotatable tokens ensure total control over who can read or publish to your library.
The problem we are solving
AI assistants do not share state. Custom instructions written for Claude on the web are invisible to Cursor on your laptop, and neither tool can access the prompt library you built inside ChatGPT. Your skills become fragmented across individual tools, forcing you to start from scratch whenever you switch devices, apps, or agent harnesses.
This fragmentation introduces friction and wasted time. Re-authoring prompts across platforms creates inconsistency and unnecessary overhead. SkillVault gives your AI skills a single, central home accessible by every agent you use.
How the vault works
SkillVault provides four unified delivery surfaces for your skill library:
1. Authenticated Vault URLs. Your vault features a base address at https://app.skillvault.dev/v/<username>. For web and mobile AI tools, short-lived or long-lived read URLs (/v/<username>/t/<token>) return clean Markdown documents formatted for LLM context windows. Dedicated publish URLs (/v/<username>/p/<token>) allow authorized agents to upload or update skills on your behalf.
2. JSON REST API. GET /api/v1/skills lists your library, and GET /api/v1/skills/<name> retrieves specific skill details. Custom applications, internal tools, and REST-capable connectors interact with your vault as structured JSON data.
3. MCP Server Surface. The native MCP endpoint (/mcp) allows Model Context Protocol clients to connect directly to your vault. MCP-compliant agents discover your available skills as tools and invoke them mid-conversation.
4. CLI Wrapper Tool. The skillvault CLI drops lightweight wrapper files into detected local harness directories. Claude Code receives a SKILL.md file, Cursor receives a .cursor/rules/skillvault.mdc file, and harnesses like Windsurf, Aider, OpenCode, Codex, Cline, Hermes, and DeepSeek Harness receive their respective native configurations.
Updating a skill in your vault automatically updates every surface. The next request from any URL, API call, MCP client, or CLI harness returns the latest version.
What lives in a skill
Each skill in your vault consists of:
- Name: A lowercase, hyphen-separated identifier unique to your vault, used in URLs, CLI commands, and API endpoints.
- Description: A concise overview that helps AI agents discover the right skill for a task and displays in your dashboard.
- Body: The Markdown instructions, prompts, playbooks, or step-by-step guides for your AI.
- Optional Files, Scripts, and Proxy Rules: Attached code assets, runtime dependencies, or proxy rules for executing API calls with stored secrets.
Skill bodies have no rigid schema restrictions. You can format them as system prompts, few-shot examples, or detailed operational workflows.
How skills reach agents
Modern AI agents possess powerful web-fetching capabilities, search MCPs, and native API connectivity. SkillVault supports all of them:
- Web and Mobile Chat AIs: Agents with browsing capabilities fetch your vault URL (
/v/<username>/t/<token>) and read skill Markdown directly into their context window. - MCP-Native Agents: Clients configured with your vault's MCP endpoint automatically query your skills as available tools during conversations.
- Local Coding Agents: Desktop tools like Claude Code, Cursor, and Windsurf read the local wrapper skill created by the CLI, directing the agent to fetch instructions from your vault.
- Custom Apps & Scripts: REST clients authenticate via
Authorization: Bearerheaders to pull skill bodies into custom prompts programmatically.
What we do not do
We do not run model inference. SkillVault stores and serves your skill library; it does not process your AI prompts or decide when a skill should be triggered. The AI model or harness you choose handles execution.
We do not track agent activity. There are no tracking scripts or model-side telemetry on your skills. The vault log records access metadata for your own review, not for advertising or external analytics.
Security and privacy
Your account password proves ownership of your SkillVault account, while rotatable vault tokens authorize API and URL access. Keeping credentials separate allows you to:
- Rotate access tokens instantly from your dashboard without changing your account password.
- Issue separate read tokens (
/t/) and short-lived publish tokens (/p/) for specific devices or agent sessions. - Revoke compromised tokens without affecting other active devices.
Skill bodies and attached assets are stored in encrypted object storage. Vault metadata is stored securely in PostgreSQL in the EU. All traffic in transit is encrypted using TLS.
We do not sell, share, or monetize your skill content. Your vault belongs entirely to you.
Pricing
SkillVault offers flexible options for individuals and teams:
- Free: 3 skills, 50 agent memories, 100 MB storage. Perfect for trying the platform with a core set of skills and agent memory.
- Pro: €8.95/month. 50 skills, 5,000 memories, 2 GB storage. Ideal for daily multi-agent workflows.
- Power: €24.95/month. Unlimited skills, 50,000 memories, 15 GB storage, neural reranking. For power users whose vault is central to their daily work.
- Team & Enterprise: €14.95/seat/month. Unlimited skills, 100k memories/org, shared team vaults, RBAC, and org governance.
All plans include full access to all four delivery surfaces (Vault URL, JSON API, MCP Server, and CLI tool).
How to start
Create your account at /auth/signup to access your dashboard and copy your tokenless vault URL (https://app.skillvault.dev/v/<username>).
To connect a coding agent (such as Claude Code, Cursor, or Codex), paste the tokenless URL into your agent's chat and ask it:
Setup SkillVault with https://app.skillvault.dev/v/<username>
Your agent will initiate the setup, opening a browser window for you to log in and approve access. Once authentication completes, your device receives its permanent credentials and downloads a local wrapper skill that makes your vault accessible across all harnesses on your device.
For web chat AIs (ChatGPT, Claude.ai, Gemini), paste your vault URL or short-lived read URL directly into your conversation.
Get in touch
Have questions, security disclosures, or enterprise inquiries? Contact us anytime at support@skillvault.dev.