Phronesitron

One-shot LLM CLI with optional MCP tool support. Uses any LLM provider (Ollama, OpenAI, Anthropic, …) via acatome-lambic.

Quick start

# Simple query (default: ollama/qwen3.5:9b)
ph explain quantum tunneling in 3 sentences

# Use a different model
ph -m anthropic/claude-sonnet-4-20250514 "summarize this" < paper.txt

# Pipe input
echo "some text" | ph rewrite this formally

# With MCP tools — the LLM can search your paper library
ph --mcp precis "what papers do we have on chlorine evolution?"

# Multiple MCP servers
ph --mcp precis --mcp perplexity "compare our MOF results to recent literature"

Install

pip install phronesitron
# or with litellm for non-Ollama providers:
pip install "phronesitron[litellm]"

For Ollama (default), make sure ollama is running locally. For OpenAI/Anthropic, set the usual env vars (OPENAI_API_KEY, ANTHROPIC_API_KEY).

Options

ph [options] <question...>

  -m, --model MODEL      provider/model (default: ollama/qwen3.5:9b)
  --mcp SERVER           enable MCP server (repeatable)
  -t, --temperature T    sampling temperature (default: 0.7)
  --max-tokens N         max output tokens (default: 4096)
  --no-think             disable reasoning mode
  -u, --unformatted      don't word-wrap output
  -s, --system PROMPT    system prompt
  --no-color             disable colored output
  -v, --verbose          show tool calls and metadata

MCP shortcuts

Built-in shortcut names for --mcp:

• precis — uvx precis-mcp (paper library)
• perplexity — uvx perplexity-sonar-mcp (web search)
• gripe — uvx gripe-mcp (issue tracking)

You can also pass a full command: --mcp "python -m my_server".