presence-mcp
PyPI launcher that exposes the presence Claude Code plugin's living project model and outcome telemetry as MCP resources for Claude Desktop, Cursor, and Continue.
presence-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
MIT. See LICENSE.
Related Servers
LAIN-mcp
Rust MCP server that gives AI coding agents architectural awareness — persistent knowledge graph, blast radius analysis, co-change detection via git, and local semantic search. No API keys, runs entirely on-premise.
Dakera
Self-hosted Rust-based MCP server for AI agent memory — persistent, queryable memory with hybrid search, knowledge graphs, built-in embeddings, and 14 core tools (expandable to 86+ with profile-based tiering).
Recall
Open-source MCP memory server for AI coding agents — durable cross-session memory, per-agent namespaces, ChromaDB-backed, self-hosted.
AgentMemory MCP
Governed persistent memory for AI agents.
BrainFlow - Shared Team Memory - MCP
Your company's brain, connected to Claude, ChatGPT, Gemini, Cursor, and VS Code. Turn your team's email history into shared memory that any AI assistant can query with natural language.
structured.sh
Self-hosted MCP server for persistent agent memory. Define typed schemas, write records, and query with SQL via DuckDB against Parquet files on disk.
mem0-mcp-server
mem0-mcp-server — exposes Mem0 persistent semantic memory as an MCP HTTP server; supports add/search/read/update/delete operations and semantic search for agent memory.
MarsNMe
Agent-agnostic persistent memory backend. 13 MCP tools, Supabase + Jina embeddings, multi-profile isolation, semantic recall across sessions.
piia-engram
Persistent AI memory across tools — remember your preferences, code standards, and decisions across Claude Code, Cursor, Codex, and any MCP tool. Local-first, zero-cloud.
Maindex Smart Memory Service
Simple, user-controlled memory for all your AI: keep, recall, update, and forget across sessions.