About presence-mcp
presence-mcp is a PyPI launcher for presence's existing MCP server, intended for users configuring MCP-aware clients. It locates a local presence installation and forwards stdio, avoiding a machine-specific script path in client configuration.
Key features
- 01Launch the local presence MCP server
- 02Expose project model and telemetry resources
- 03Use one command in client configurations
- 04Override CLI and Python executable paths
Use cases
- 01Configure presence for Claude Desktop
- 02Connect Cursor or Continue to presence
- 03Select a project-specific working directory
- 04Run presence with a separate Python interpreter
MCP support
The package itself does not implement MCP; it locates presence and forwards stdio to its MCP server. That server exposes two read-only resources per repository: a living project model and recent outcome telemetry.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMEpresence-mcp
PyPI launcher that exposes presence's living project model and outcome telemetry as MCP resources for Claude Desktop, Cursor, Continue, and other MCP-aware clients.
presence is a Claude Code plugin. It already ships a stdio MCP server at lib/cli.py mcp and exposes two read-only resources per repository (see What you get). This package is the PyPI shim that lets that server be listed in the official MCP Registry and invoked with a single command name in every per-client config, instead of an absolute path that varies per machine.
The launcher itself speaks no MCP. It locates a local presence install and forwards stdio to python lib/cli.py mcp. About 40 lines of Python, stdlib only.
Quick start
pip install presence-mcp
Then point any MCP client at the presence-mcp command. For Claude Desktop, that's:
{
"mcpServers": {
"presence": {
"command": "presence-mcp"
}
}
}
Full per-client config for Cursor and Continue below. Requires Python 3.12+ and a local presence install.
What you get
Two read-only MCP resources, one per repository you've worked on:
| URI | Content | MIME type |
|---|---|---|
presence://<repo_id>/model | Living model.md (Markdown) | text/markdown |
presence://<repo_id>/telemetry | Recent commit / revert / verification claims (JSON array) | application/json |
<repo_id> is a 12-char SHA-256 prefix of the repo. The server resolves the current repo from the launcher's working directory; see Working-directory caveat. Resource schema and protocol details: presence's docs/mcp.md.
Per-client configuration
The same presence-mcp command works in every client.
Claude Desktop
~/Library/Application Support/Claude/claude_desktop_config.json (macOS), or the equivalent on Windows / Linux:
{
"mcpServers": {
"presence": {
"command": "presence-mcp"
}
}
}
Cursor
.cursor/mcp.json in the project root, or the global Cursor MCP settings:
{
"mcpServers": {
"presence": {
"command": "presence-mcp"
}
}
}
Continue
~/.continue/config.json:
{
"mcpServers": [
{
"name": "presence",
"command": "presence-mcp"
}
]
}
Verify the install
The launcher accepts JSON-RPC on stdin. Pasting {"jsonrpc":"2.0","id":1,"method":"initialize"} and pressing Enter should yield an initialize response. You can also invoke it as python -m presence_mcp if a console script isn't convenient.
Environment variables
| Variable | Purpose |
|---|---|
PRESENCE_MCP_CLI | Absolute path to presence's lib/cli.py. Overrides the default ~/.claude/plugins/presence/lib/cli.py lookup. If set but the path is missing, the launcher errors out instead of falling back, so typos surface. |
PRESENCE_MCP_PYTHON | Python interpreter to run presence with. Defaults to sys.executable of the launcher itself. Useful when presence is installed under a different Python (for example, a project venv). |
If presence-mcp reports it could not locate presence, install presence at the standard Claude Code plugin path, or set PRESENCE_MCP_CLI.
Working-directory caveat
presence's MCP server resolves the current repo from the launcher's working directory via git rev-parse --show-toplevel. MCP clients that launch the server from a fixed directory will only see one repo. To switch repos, either launch a separate instance per project (set cwd in the client's per-server config), or restart the server from the target project root.
Compatibility
- Python: 3.12 or newer. CI matrix covers 3.12 / 3.13 / 3.14.
- presence: tested with
>= 0.6.0. Older versions back to v0.4.1 (where the MCP server first shipped) may still work but are not exercised by CI. - Platforms: Linux and macOS exercised in CI. Windows is untested in v0.1.0. The launcher uses
subprocess.runwhich is cross-platform, but the default presence install path (~/.claude/plugins/presence/lib/cli.py) needs a real Windows install to confirm. Report issues if you try it.
This launcher is protocol-agnostic: it does not rev when presence ships new MCP resources. It only revs when its own resolution logic, CLI surface, or server.json content changes.
Where to file issues
- Launcher bugs (path detection, install errors, CLI argument forwarding): issues on this repo.
- Resource-content bugs (what
presence://*/modelreturns, telemetry schema, working-directory behavior, anything about the protocol itself): issues on the presence repo.
Disambiguation
presence's MCP server advertises serverInfo.name = "presence-mcp" in its initialize response. That's the server's MCP identity. This package is the PyPI launcher. The names match by intent.
Links
- PyPI: https://pypi.org/project/presence-mcp/
- MCP Registry: https://registry.modelcontextprotocol.io/v0/servers?search=presence-mcp (
io.github.sara-star-quant/presence-mcp) - Upstream (presence): https://github.com/sara-star-quant/presence
License
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.