Account
Sign in or create an account to connect your AI.
Sign in
MCP (Model Context Protocol) lets AI agents use fraglets as tools. Add this config to any MCP-compatible client — Claude Desktop, Claude Code, OpenClaw, or whatever you use.
Copy this JSON into your MCP client settings:
Loading...This token is included in the config above.
Loading...
Paste this into a terminal to confirm your token works. It lists the available MCP tools and returns immediately.
Loading...Having trouble? Client-specific notes
The token must be valid. fraglet.com does not run an OAuth authorization server. As long as the bearer token in your config is current, the initial request succeeds and no OAuth flow runs. If the token is missing, empty, or expired, the client falls into an OAuth re-auth path and will fail — regenerate your token on this page if you see "Invalid OAuth error response" or a 401 loop.
Shell escaping gotchas. If you're passing the token via a shell substitution like $(cat ~/.secrets/mcp_token), make sure your outer quoting lets the substitution expand. A common failure is a literal "Bearer " header with nothing after it — that means the token never interpolated.
Direct Streamable HTTP clients: point at https://fraglet.com/mcp/ with Authorization: Bearer <your token>. Protocol version 2025-06-18.
Discovery: a fresh client can hit GET /mcp/info (no auth required) to learn the transport, auth model, and this page's URL in a single JSON document.
Services that use your fraglets
These services accept fraglets for personalised recommendations. Add them as MCP servers alongside fraglets.
London restaurant and live music recommendations matched to your food and music fraglets. Supports individual and group recommendations.
food music
Set up another agent
Paste this to any AI agent to get it set up with fraglets:
Read https://fraglet.com/skill.md and follow the setup instructions. I already have an account — here's my MCP config:Then paste your MCP config (above) after this message. The agent will read the skill guide and configure itself.
Building an agent?
Your agent has a cold-start problem. Every new user interaction starts from zero. Fraglets solve it: structured, user-curated preference profiles your agent can read via MCP. No 20-question interview, no guessing.
Available tools: list_my_fraglets, select_fraglets (picks the right subset for your intent),
discover_fraglets, list_my_jackets, and more.
Full docs at fraglet.org
Next steps
Import or create fraglets. The fastest way is to import what your AI already knows about you.
View your fraglets to manage, review, and use the picker.
Compose jackets — personas built from any combination of your own and others' fraglets.