MCP Prompt Manager
Tae4anLABSCO SUMMARY
Prompts as files, watched — so editing one is the whole workflow.
The live monitoring is what makes this more than a loader. Saving a markdown file and having it immediately available in the CLI turns prompt work into ordinary editing, and the notification means you are not restarting a session to pick up a fix. Because prompts are just files in directories you name, they version-control and share like anything else in a repository. It is built for Amazon Q's CLI specifically, and it exposes prompts rather than tools — which is the right shape for what it does.
What it is
A single-file MCP server for the Amazon Q Developer CLI that serves your *.md files as MCP prompts. It watches the configured directories, so adding, editing or deleting a prompt file updates what Q offers without restarting anything. It exposes prompts rather than tools — the prompt list and prompt fetch are the whole protocol surface.
What you get
- Every
*.mdfile in the configured directories offered as a prompt, listed in the CLI with/promptsand used by name - Placeholder substitution: write
{variable}in the file and supply values when the prompt is fetched, so one file covers a family of requests - Live file monitoring with an MCP notification sent when the list changes, so Q refreshes on its own
- Directories you choose, PATH-style:
PROMPTS_PATHtakes a colon-separated list on Unix and semicolons on Windows, defaulting to the Amazon Q prompts directory, which is created for you - Debug logging when you need it:
MCP_DEBUG_LOGGINGturns on full request and response tracing to a log file you can tail, withMCP_LOG_FILEchoosing the path — off by default, andMCP_LOG_LEVELsits atWARNINGotherwise
Requirements
No account and no key — it reads local files. Pure Python with no external dependencies. Run it with uvx prompt-mcp-server, or point the client's command at the server script directly; the published configuration uses uvx with the package pinned to latest. The pyproject name is prompt-mcp-server (2.0.9). You can check it outside the CLI by piping a JSON-RPC request into the script — prompts/list returns the catalogue and prompts/get returns one prompt with its arguments filled in.
Setup effort
One command — uvx prompt-mcp-server
USE CASES
01
Keeping your reusable prompts somewhere you can edit them
Write markdown files in a directory, and they show up as prompts you can call by name.
02
Making one prompt cover many cases
Use placeholders for the language, function name or description and supply them when the prompt is fetched.
03
Sharing a prompt library with a team
Point the path variable at a checked-out repository so everyone gets the same prompts and updates land by pulling.
04
Working out why a prompt is not appearing
Turn on debug logging, tail the log file, and watch the requests, responses and file-monitoring activity as you save a file.
BEFORE YOU INSTALL
What to know first
Tools you get
0 tools when we connected.
Requires
- uvx on PATH
- Python 3.6+
Where it runs
Both documented setups run the server on your machine.
INSTALLATION
Connect your client
Server setup
TERMINAL
uvx prompt-mcp-serverMCP Clients
Amazon Q — uvx
AMAZON Q — UVX · JSON
{
"mcpServers": {
"prompt-server": {
"command": "uvx",
"args": [
"prompt-mcp-server@latest"
],
"disabled": false,
"autoApprove": []
}
}
}Amazon Q — Local build
AMAZON Q — LOCAL BUILD · JSON
{
"mcpServers": {
"prompt-server": {
"command": "python3",
"args": [
"mcp_server/prompt_mcp_server.py"
],
"timeout": 10000
}
}
}
