Borzoi Remote Tool (MCP Server)#
TOU validation and deployment status (2026-08-16)#
A fresh Python 3.12 service environment resolved borzoi-pytorch 0.4.4 with supported Transformers 4.50.3, loaded official weights on the GB10, and returned bounded finite prediction and variant-effect results through a strict loopback TOU endpoint. Four concurrent fixture calls passed with serialized model access. NVIDIA’s aarch64 cusparselt wheel still reports incompatible platform metadata in
uv pip check; public publication, saturation, recovery, cross-user isolation, and biological accuracy remain incomplete. Authenticated private Platform import and owner testing passed on 2026-08-16; public publication and independent-caller authorization/isolation remain untested.
Operations:
run_borzoi_predict,run_borzoi_variant_effectStart:
python -m tooluniverse.remote.borzoi.borzoi_toolEndpoint:
http://127.0.0.1:8012/mcpProvider configuration: keep
HF_HOMEandTORCH_HOMEunder provider-owned caches. Install this directory’s requirements in an isolated environment; Borzoi must usetransformers>=4.46,<4.51.TOU check:
tu doctor --forward http://127.0.0.1:8012/mcp --jsonPrivate relay:
tu serve --share --forward http://127.0.0.1:8012/mcp --name validation-borzoi --workers 1
Non-loopback binding requires TOOLUNIVERSE_API_TOKEN; otherwise keep the server on loopback. The relay requires TOOLUNIVERSE_SERVICE_KEY. Public publication and independent-caller testing were not run.
New-user check: python scripts/remote_validation/setup_skill_preflight.py --implementation borzoi. Add --check-provider-env before launch, --live after launch, and --check-connect-prereqs before sharing. Live preflight checks exact MCP discovery only; it does not run or validate a model. The pinned relay SDK is not on PyPI and currently requires authorized GitHub repository access plus a configured SSH key, so a working local MCP server does not by itself prove that a new operator can share it.
The authenticated 2026-08-16 Platform matrix found all 30 private owner relays online and all 41 operations discoverable. This implementation was imported as unpublished owner draft(s), configured with a 120-second timeout, and invoked through /expert-sessions/{id}/test. Across the set, 38 unique operations passed return-schema and semantic validation; the three USPTO operations returned exact provider HTTP 403 and remain credential-blocked. Public publication, independent-caller authorization/isolation, broad saturation, and persistent supervision were not tested.
See the complete setup and verification guide.
Serves Borzoi (Linder et al., Nature Genetics 2025) — sequence → RNA-seq/multi-omic coverage prediction, the successor to Enformer — as ToolUniverse remote tools: run_borzoi_predict and run_borzoi_variant_effect.
Served remotely (not bundled) because it needs PyTorch + ~0.8 GB of weights per replicate (johahi/borzoi-replicate-{0..3}) and is GPU-recommended.
Deploy#
pip install -r requirements.txt # torch + borzoi-pytorch
python borzoi_tool.py # starts the MCP server on 127.0.0.1:8012
Expose remotely only behind TOOLUNIVERSE_API_TOKEN (SMCP bind guard).
Register in ToolUniverse#
Tool definitions: src/tooluniverse/data/remote_tools/borzoi_tools.json
(type: RemoteTool). Connect via the standard MCPAutoLoaderTool/server_url
mechanism. First call lazily downloads and caches the weights.