Claude-Mem
Persistent memory plugin for Claude Code and other coding agents, with MCP tools to search past sessions
Serves a bundled library of about 1,300 markdown work skills and workflow recipes to MCP clients
By Mohit Aggarwal (individual) JavaScript Remote (hosted) · Streamable HTTPLocal · stdio Knowledge and memoryProductivity
75 popularity score ★ 1.5k ↓ 2.1k/wk
Opens the app, which asks before it adds the server.
Its docs show no set-up for a particular client: add this address in your client the way it adds a remote MCP server.
https://pm-skills-mcp.pm-claude-skills.workers.devOpens the app, which asks before it adds the server.
claude mcp add pm-claude-skills -- npx -y -p pm-claude-skills pm-claude-skills-mcp{
"mcpServers": {
"pm-claude-skills": {
"command": "npx",
"args": [
"-y",
"-p",
"pm-claude-skills",
"pm-claude-skills-mcp"
]
}
}
}{
"mcpServers": {
"pm-claude-skills": {
"command": "npx",
"args": [
"-y",
"-p",
"pm-claude-skills",
"pm-claude-skills-mcp"
]
}
}
}{
"servers": {
"pm-claude-skills": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"-p",
"pm-claude-skills",
"pm-claude-skills-mcp"
]
}
}
}[mcp_servers.pm-claude-skills]
command = "npx"
args = ["-y", "-p", "pm-claude-skills", "pm-claude-skills-mcp"]gemini mcp add --scope user pm-claude-skills npx -- -y -p pm-claude-skills pm-claude-skills-mcpMCP_APPS |
Optional: Set to 1 to serve the experimental embedded skill-picker resource (MCP Apps draft). |
|---|
pm-claude-skills is mainly a collection of SKILL.md files. Each one is a plain-markdown method for a professional task, such as writing a PRD, scoring a backlog with RICE or reviewing a lease, and comes with an output template and quality checks. The npm package ships two programs. `pm-claude-skills` is an installer CLI that copies skills into an AI tool's config folder. `pm-claude-skills-mcp` is the MCP server: it hands the same library to any MCP client when asked, so nothing needs copying in advance. At 83.2.0 the package description counts 1,310 skills.
The local server talks over stdio and uses only the Node standard library. It reads the skills folder bundled in the package and never goes online. It has 9 tools. They list the library or filter it by maturity tier, search it by keyword, return one skill's full text or its declared input fields, and say which of two similar-looking skills to pick (and whether a name has been retired). Two more list and fetch workflow recipes, which chain several skills in order. `run_skill` uses MCP sampling to have the client's own model write the finished output, so the server needs no model key. `check_contrast` works out WCAG 2.1 and APCA contrast for two hex colours with plain arithmetic. The source also publishes every skill as an MCP prompt (named by skill id) and as a `skill://<name>` resource, which the README does not cover.
The maker also runs a hosted copy on Cloudflare Workers. As of October 2026 the project's GitHub repository, its account and its GitHub Pages site all return 404. The npm package is the only place left to get the code.
| To run it yourself | Node.js 18+ |
|---|---|
| Runs on | macOS, Windows, Linux |
Run it with npx; nothing needs installing first. No account or API key is needed. `run_skill` only produces output in clients that support MCP sampling.
The local server in v82.0.0 has 9 tools (mcp/server.mjs); the three render_* widget tools exist only on the hosted Worker. `search_skills` returns 10 results by default, up to 50 with `limit`. `run_skill` falls back to returning a message when the client does not support sampling. An interactive skill-picker UI is marked experimental and is off unless `MCP_APPS=1` is set. The skills are guidance text: the server does not check the quality of what the model writes from them.
Good for anyone who wants an assistant to pull product-management, design or other work methods on demand, without copying hundreds of skill files into each tool. The local server is read-only, works offline and needs no sign-in. If you use one tool that loads SKILL.md files natively, the package's own `add` installer may be the simpler route. With the repository gone, there is nowhere to report a bug or read the source except the published package, so weigh that before relying on it. The hosted endpoint was not serving its catalogue when checked, so use the local npx form. Pick something else if you need a server that acts on your project tools.
9 only read. A ? marks an optional parameter.
list_skillstier?: enumsearch_skillsquery: stringlimit?: integerdisambiguate_skillname: stringget_skillname: stringget_skill_inputsname: stringrun_skillname: stringinput?: stringinputs?: objectlist_workflowsget_workflowid: stringcheck_contrastforeground: stringbackground: stringlevel?: enumskill://<name>ui://pm-skills/picker<skill id>