Your prompt library,
inside every AI tool
The PromptCase MCP server connects the prompts you've captured everywhere to Claude Code, Cursor, VS Code, Windsurf, Codex and any MCP-capable app — so your AI can search them, reuse them, improve them and save new ones, mid-conversation.
One page, one copy-paste per app. Free on every plan.
Works with
One connection, every tool
MCP (Model Context Protocol) is the open standard AI apps use to reach outside tools. PromptCase speaks it natively: connect once per app and your whole library — captured from ChatGPT, Claude, your terminal, everywhere — is right there in the conversation.
search_prompts
Fast search over your own library — the AI finds what you already wrote.
get_prompt
Pulls the full text of any prompt into the chat.
save_prompt
Saves a good prompt mid-session — honoring your capture switch and secret redaction.
enhance_prompt
Improves a draft with PromptCase AI, in your own style (needs AI Brain enabled).
list_folders
Sees how your library is organised.
Favourites become slash commands
Star a prompt in your dashboard and it shows up as a native slash command in Claude Code, Cursor and VS Code — your 20 newest favourites, always in sync. Type /, pick it, and the full saved prompt drops into the conversation exactly as you wrote it.
No copy-paste, no hunting through tabs — the prompt you refined last month is one keystroke away in today's session.
Install in two steps
Server URL: https://promptcase.dev/mcp · authentication is a single token in a header — no OAuth, no app registration.
Step 1 — get your setup code (once)
Sign in and open Dashboard → Devices → MCP Server. Name the connection (e.g. "Claude Code"), click Generate setup code, pick your app and copy the ready-made snippet — the token is already inside it. Tokens last ~180 days; mint one per app so you can revoke them individually.
Step 2 — paste it into your app
The setup-code page generates each of these with your token filled in. Replace pc_tok_YOUR_TOKENif you're writing the config by hand.
Claude Code — one command
claude mcp add promptcase --scope user --transport http https://promptcase.dev/mcp --header "Authorization: Bearer pc_tok_YOUR_TOKEN"
Claude Desktop
Settings → Developer → Edit Config (claude_desktop_config.json), then restart. The header is split across args/env deliberately — a known Windows bug mangles spaces in args.
{
"mcpServers": {
"promptcase": {
"command": "npx",
"args": ["mcp-remote", "https://promptcase.dev/mcp", "--header", "Authorization:${AUTH_HEADER}"],
"env": { "AUTH_HEADER": "Bearer pc_tok_YOUR_TOKEN" }
}
}
}Cursor
~/.cursor/mcp.json (global) or .cursor/mcp.json (per project).
{
"mcpServers": {
"promptcase": {
"url": "https://promptcase.dev/mcp",
"headers": { "Authorization": "Bearer pc_tok_YOUR_TOKEN" }
}
}
}VS Code
Create .vscode/mcp.json in your project (top-level key is servers). This connects VS Code as an MCP client — for the PromptCase sidebar and prompt palette inside the editor, that's the separate VS Code extension.
{
"servers": {
"promptcase": {
"type": "http",
"url": "https://promptcase.dev/mcp",
"headers": { "Authorization": "Bearer pc_tok_YOUR_TOKEN" }
}
}
}Windsurf
~/.codeium/windsurf/mcp_config.json — note the key is serverUrl.
{
"mcpServers": {
"promptcase": {
"serverUrl": "https://promptcase.dev/mcp",
"headers": { "Authorization": "Bearer pc_tok_YOUR_TOKEN" }
}
}
}OpenAI Codex CLI
~/.codex/config.toml, with the variable exported in your shell profile. Codex supports the tools; it doesn't render slash commands yet.
# ~/.codex/config.toml [mcp_servers.promptcase] url = "https://promptcase.dev/mcp" bearer_token_env_var = "PROMPTCASE_TOKEN" # shell profile (e.g. ~/.zshrc) export PROMPTCASE_TOKEN="pc_tok_YOUR_TOKEN"
Gemini CLI
~/.gemini/settings.json — note: httpUrl, not url.
{
"mcpServers": {
"promptcase": {
"httpUrl": "https://promptcase.dev/mcp",
"headers": { "Authorization": "Bearer pc_tok_YOUR_TOKEN" }
}
}
}Get the most out of it
Connecting is step one. The teams that get real leverage out of PromptCase run their library as a loop — and teach their AI to run it for them.
Search before you write
Say "check my PromptCase for a prompt about API design reviews first". The best prompt is usually the one you already refined three sessions ago.
Star your daily drivers
Favourites become slash commands. Keep the 10–20 prompts you reach for weekly starred, and they're one keystroke away in every tool.
Save the winners as you go
When a prompt lands, say "save that to my PromptCase as ‘Bug triage brief’". Ten seconds now, reusable forever — and labeled "MCP" in your dashboard.
Enhance instead of rewriting
"Enhance this with PromptCase" rewrites a rough draft in your own voice, informed by your history — then save the improved version back.
Teach your AI to use it — paste this once
AI agents use tools far more when you tell them when to. Paste this into your project's CLAUDE.md, AGENTS.md, or your editor's rules file, and your AI will search your library before improvising, and offer to bank the good prompts it helps you write:
## PromptCase (my prompt library via MCP) You have access to my PromptCase prompt library through the promptcase MCP server. Use it proactively: - Before writing a substantial prompt, brief, or reusable instruction from scratch, call search_prompts first — I may already have a refined version. Prefer reusing my saved prompt over improvising. - When a result looks relevant but truncated, call get_prompt with its id to read the full text before using it. - When we craft a prompt, brief, or checklist worth keeping, offer to save it with save_prompt (short descriptive title). Don't save trivial one-offs or anything containing secrets. - When I ask you to improve a prompt, use enhance_prompt — it rewrites in MY style using my prompt history. - Use list_folders when you need to know how my library is organised.
Works verbatim in Claude Code (CLAUDE.md), Codex (AGENTS.md), Cursor rules and Windsurf rules. Adjust the tone to taste — the pattern that matters is search first, save the winners.
Troubleshooting
- 401 / "Invalid or expired token" — the token was revoked or is past its ~180 days. Generate a fresh setup code from the MCP page and update the config. Make sure the header value is the bare
pc_tok_…string — not a JSON object. - "AI features are not enabled on this account" — that's
enhance_promptonly; enable AI Brain in your dashboard. Every other tool works without it. - "Capture is turned off — nothing was saved" — you switched capture off in PromptCase Settings; flip it back on to save from MCP again.
- Plan limits — search and reads only ever see the prompts your plan makes visible, exactly like the dashboard.
Two minutes to connect
Your prompts are already in PromptCase. Put them to work everywhere you build.
Get your setup code