Computase

Computase is a local Python library for small, well-defined DNA and RNA sequence calculations:

• nucleotide composition, GC bounds, and GC skew
• DNA or RNA reverse complements
• translation with selectable NCBI genetic-code tables
• six-frame candidate ORF enumeration
• IUPAC motif searches on either strand

The Python API is the primary interface. The same operations are also available through the optional Model Context Protocol (MCP) interface for local agent workflows. Computation runs locally; input sequences are not sent to a service.

Install

For most Python environments:

pip install computase

For a project managed with uv:

uv add computase

Python 3.11 or newer is required.

Quick start

from computase.seq import translate_sequence

sequence = ">synthetic-cds\nATGGCCATTGTAATGGGCCGCTGAAAGGGTGCCCGATAG\n"
result = translate_sequence(sequence, table_id=1)

print(result.model_dump())

Representative output (computase_version matches the installed release):

{'computase_version': '<installed version>', 'parameters': {'table_id': 1, 'stop_handling': 'translate-through'}, 'protein': 'MAIVMGR*KGAR*', 'table_id': 1, 'table_name': 'Standard', 'stop_handling': 'translate-through', 'codon_count': 13, 'stopped_early': False}

Results are typed Pydantic models. They record the Computase version and effective parameters, but never echo the full input sequence.

See the Python examples for concise, runnable examples of all five sequence operations.

Capabilities and validation

These checks establish the documented conventions and regression boundaries; they do not establish correctness for every biological interpretation or use case. If a result differs from an independent reference, use the scientific correctness report with a minimized, non-sensitive sequence.

Scientific scope and conventions

• Inputs are raw nucleotide strings or a single FASTA record, not multi-record files.
• Coordinates are 0-based and end-exclusive on the normalized forward reference, after FASTA headers and whitespace are removed.
• Strand is reported separately; normalized_sequence[start:end] reproduces each reported forward span.
• ORFs are sequence candidates, not gene predictions.
• IUPAC GC bounds preserve uncertainty rather than assigning probabilities.
• Sequence length is capped at 5,000,000 nucleotides; motif and result limits are enforced.
• Computase 0.1.x does not fetch records, align sequences, or annotate genes.

Optional MCP interface

stdio

With uv installed, uvx can run the MCP server without installing Computase into the current environment:

{
  "mcpServers": {
    "computase": {
      "command": "uvx",
      "args": ["computase"]
    }
  }
}

If Computase was installed with pip into an environment available to the MCP client, use computase as the command and omit the arguments. For a uv-managed project, run uv run computase from the project root; configure the MCP client with uv as the command and ["run", "computase"] as the arguments.

The five tools are computase_summarize_sequence, computase_reverse_complement, computase_translate_sequence, computase_enumerate_orfs, and computase_scan_motif.

Streamable HTTP

uvx computase --transport streamable-http --host 127.0.0.1 --port 8000

Connect an MCP client to http://127.0.0.1:8000/mcp. HTTP binds to localhost by default.

Do not expose the Computase HTTP server directly to a public network. Non-loopback deployment requires a separately managed TLS boundary that authenticates every request and enforces request-size, concurrency, and rate limits.

Companion Skill

The repository and source distribution include a Computase companion Skill that teaches agents when and how to choose the five operations. Prefer the Computase MCP tools when configured; otherwise the Skill can run an isolated uv fallback that resolves Computase without mutating the active environment. Its usage examples cover MCP, runner, and Python workflows. These Skill files are not installed by the Python wheel.

Install the Skill into a supported agent with:

npx skills add madhusudan-kulkarni/computase --skill computase

skills.sh indexes public repositories from install telemetry. There is no separate submission form.

Development

Use uv sync --locked --extra dev, then run:

uv lock --check
uv run --locked ruff format --check src tests evaluations scripts skills/computase/scripts
uv run --locked ruff check src tests evaluations scripts skills/computase/scripts
uv run --locked mypy src tests evaluations scripts skills/computase/scripts
uv run --locked pytest -q
uv run --locked python -m evaluations.runner

See CONTRIBUTING.md for reference-vector requirements.

Roadmap

See ROADMAP.md for planned capabilities, including melting temperature calculations, restriction digestions, and a client-side WebAssembly interface.

Citation

If Computase contributes to your work, cite the software metadata in CITATION.cff. GitHub also exposes this through Cite this repository. Computase is also registered on bio.tools.

License

Computase is licensed under the MIT License.