Computase

Verified local bioinformatics computation.

Computase provides five sequence operations through equally supported Python and MCP interfaces:

• sequence composition, GC bounds, and GC skew
• DNA/RNA reverse complements
• translation with selectable NCBI genetic codes
• six-frame candidate ORF enumeration
• IUPAC motif scanning on either strand

Computase is sequence-only in 0.1.0. It does not predict genes, fetch remote records, run alignments, or provide generic statistics.

Install

pip install computase

Python 3.11 or newer is required. Computation is local; input sequences are not sent to a service.

Python

from computase.seq import reverse_complement, summarize_sequence

summary = summarize_sequence(">example\nAAGCSWN\n")
print(summary.gc_min_percent, summary.gc_max_percent)

result = reverse_complement("AUGC")
print(result.reverse_complement)

All operations accept raw sequences or one FASTA record. Results include the Computase version and effective parameters but never echo the full input sequence.

MCP over stdio

{
  "mcpServers": {
    "computase": {
      "command": "computase"
    }
  }
}

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

MCP over streamable HTTP

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.

Scientific conventions

• Coordinates are 0-based and end-exclusive on the forward reference.
• Coordinates index the normalized sequence after FASTA headers and whitespace are removed.
• Strand is reported separately; normalized_sequence[start:end] reproduces each forward span.
• ORFs are 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.

Companion Skill

skills/computase/SKILL.md teaches agents when and how to choose the five MCP tools. Longer examples are in skills/computase/references/usage-examples.md.

Development

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

uv run ruff format --check src tests evaluations
uv run ruff check src tests evaluations
uv run mypy src tests evaluations
uv run pytest -q
uv run python -m evaluations.runner

See CONTRIBUTING.md for reference-vector requirements.

License

MIT. See LICENSE.