teasel

Connect lab instruments to AI assistants.

teasel is an open-source platform that lets large language models talk directly to bench instruments — oscilloscopes, function generators, multimeters, power supplies — through the Model Context Protocol (MCP).

teasel.tools — browse supported instruments

What's in this repo

Package	PyPI	Description
teasel-cli/	teasel	CLI and TUI for configuring which instruments are available
teasel-server/	teasel-server	MCP server that exposes instrument tools to AI assistants

How it works

1. Run uvx teasel to browse the instrument registry, enter connection details, and generate a .mcp.json config file.
2. Claude Code (or any MCP-compatible assistant) reads .mcp.json and starts teasel-server automatically at the beginning of each session.
3. The AI can now control and measure with your instruments directly.

Instruments

Bundled in teasel-server:

• LeCroy WaveSurfer — oscilloscope (Ethernet / USB)
• Philips PM5190 — LF function generator (GPIB via AR488)

Community instruments are listed in the instruments registry.

Third-party drivers

Any Python package can add instruments to teasel-server by registering a teasel.instruments entry point. No changes to this repo required.

Running with a local LLM

teasel local connects teasel-server to a local LLM via llama-server — no Claude or Anthropic API needed.

1. Get llama-server and a model

pip install llama-cpp-python[server] huggingface-hub
huggingface-cli download bartowski/Ministral-3B-Instruct-2410-GGUF \
  --include "Ministral-3B-Instruct-2410-Q8_0.gguf" --local-dir ~/models

2. Start llama-server

Local (CPU-only):

llama-server -m ~/models/Ministral-3B-Instruct-2410-Q8_0.gguf --jinja -ngl 0 --port 8080

Rented GPU (RunPod / vast.ai — recommended for demos):

On the pod, run the same install + download commands above, then:

llama-server -m ~/models/Ministral-3B-Instruct-2410-Q8_0.gguf --jinja -ngl 999 --port 8080 --host 0.0.0.0

Expose TCP port 8080 in the pod settings to get a public IP and port.

3. Start teasel-server in HTTP mode

uv run teasel-server --config teasel.toml --setup setup.toml --transport streamable-http --port 7891

4. Run the bridge

# Local llama-server:
uv run teasel local

# Rented GPU:
uv run teasel local --llama-url http://<pod-ip>:<port>

---

Development

This repo is a uv workspace. Both packages share a single venv.

Setup:

uv sync --all-packages

Run:

uv run teasel         # CLI / TUI
uv run teasel web     # Web UI (http://127.0.0.1:7890)
uv run teasel-server  # MCP server

Test against the local instruments registry instead of GitHub:

TEASEL_REGISTRY=/path/to/instruments uv run teasel

Release: Bump version in the relevant pyproject.toml and push to main. CI publishes that package to PyPI automatically via trusted publishing. Each package is released independently.

License

AGPL v3. See LICENSE.