# promptlike-mcp [Health: Active]

**Category:** 🗣️ Conversational AI  
**Repository:** https://github.com/zeynepaslierhann/promptlike-plugin  
**GitHub Stars:** 1  
**npm Downloads (last month):** 1822  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/promptlike-mcp

## Description
Your PromptLikeEngineer prompt library as MCP tools, slash prompts and resources. Add-only saves.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "promptlike-mcp": {
    "command": "npx",
    "args": ["-y","promptlike-mcp@0.7.2"],
    "env": {
      "PROMPTLIKE_API_URL": "",
      "PROMPTLIKE_CONFIG_DIR": "",
      "PLE_SUGGEST": ""
    }
  }
}
```

**Requires environment variables:** `PROMPTLIKE_API_URL`, `PROMPTLIKE_CONFIG_DIR`, `PLE_SUGGEST` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What promptlike-mcp does

promptlike-mcp brings a PromptLikeEngineer prompt library into Claude Code. Claude can find a prompt by name or search term, retrieve its content, fill its `{{variables}}`, and use the result in a conversation. The plugin also turns saved prompts into selectable slash commands and `@` resources, so a prompt can be invoked directly rather than retrieved through a conversational request.

Prompt saving is add-only. Creating a prompt does not remove existing content, updating creates another version while retaining the older one, and moving changes a prompt's repository or folder without deleting it. Saving the same content twice has no effect. Text that resembles an API key or token is rejected, and account plan limits still apply.

## How it works

Claude Code starts the MCP process through `npx`. The server runs locally and communicates with PromptLikeEngineer at `https://promptlikeengineer.com`, or with the endpoint selected through `PROMPTLIKE_API_URL`. Authentication comes from the session created by `ple login` and is refreshed through the same service when needed.

Saved prompts are available as MCP prompts and resources. A title such as “Code Review” becomes a `code-review` slash-command name, and arguments fill variables in the order those variables first appear. Variables omitted from a slash-command invocation remain in the resulting text. Resource references use the `promptlike://<repo>/<prompt>` form.

Before each Claude Code message, the optional suggestion hook runs a local match against cached prompts. It ignores slash commands, short messages, weak matches, and prompts already suggested in the session. If a match is found, Claude receives a short note naming up to three prompts and decides whether to fetch one. A background cache refresh may download the user's prompt list when the local copy is stale.

## Setup and configuration

Install the plugin from its marketplace or run Claude Code with a clone supplied through `--plugin-dir`. Install the PromptLikeEngineer CLI globally, then sign in once with `ple login`. The plugin can load without authentication, but tool calls return a login instruction instead of prompt content.

The suggestion hook requires the CLI. It can be disabled from Claude Code's configuration, with `ple config set suggest off`, or by setting `PLE_SUGGEST=off`. PromptLikeEngineer CLI files are stored under `~/.config/promptlike/` by default; `PROMPTLIKE_CONFIG_DIR` changes that location. When signed out, the hook reads from `~/.promptlike`.

## Tools and capabilities

- `list_prompts` and `search_prompts` find library entries.
- `get_prompt` retrieves a selected prompt.
- `fill_variables` substitutes values into prompt text.
- `suggest_prompts` matches a task description against the library.
- `save_prompt`, `update_prompt`, and `move_prompt` manage prompts without deletion.
- `submit_feedback` sends a short, confirmed note to the maker.
- Saved prompts appear as slash commands and `promptlike://` resources in Claude Code.

Fetching a prompt through a tool, slash command, or resource records a use event containing the prompt and version, access method, suggestion status, and MCP client identity.

## Limitations and notes

The suggestion hook performs matching locally and does not send the message itself. The MCP server does send prompt reads, requested task text for `suggest_prompts`, saved text, and confirmed feedback to PromptLikeEngineer as needed. The `suggest_prompts` task description is limited to 2,000 characters and is not logged or stored by that endpoint.

The local cache holds up to 500 prompts with metadata and the first 600 characters of each text. Feedback can be submitted without a session, but it must be a short message and cannot resemble a key or token. The server does not delete prompts; moving, updating, and saving are non-destructive operations.

_Full upstream README: https://allmcps.com/mcp/promptlike-mcp/readme_