Install Genomi without an agent by cloning the repository, installing libraries, registering MCP, and verifying the harness.

Use manual setup when you want to control every command yourself. The agent setup flow is still the canonical source of truth; this page mirrors the current installer behavior.

Prerequisites

Python 3.10 or newer.
git.
python3 -m pip, or uv for the managed-Python fallback.
macOS or Linux. On Windows, use WSL.
Java on PATH when running PharmCAT.
samtools and bcftools for BAM-derived VCF parsing.

Clone and install

export GENOMI_HOME=~/.genomi
git clone git@github.com:exon-research/genomi.git "$GENOMI_HOME/genomi"
cd "$GENOMI_HOME/genomi"
python3 scripts/install_for_agents.py --libraries everything

The installer creates:

<GENOMI_HOME>/bin/genomi - stable command shim.
<GENOMI_HOME>/reference - public reference resources.
<GENOMI_HOME>/tools - managed tool downloads such as PharmCAT and aligners.
<GENOMI_HOME>/registry.json - user/profile and Active Genome Index metadata.
<GENOMI_HOME>/sessions/.../context.json - session context.
<GENOMI_HOME>/shared-evidence.sqlite - shared reviewed evidence storage.

Add the shim to PATH if you want the command available everywhere:

export PATH="$GENOMI_HOME/bin:$PATH"

Choose a smaller library purpose

everything installs the default public libraries, roughly 7 GB. Smaller purposes are available:

Purpose	Installs
`setup-only`	No public libraries
`common-questions`	ClinVar GRCh38, HPO, GenCC
`medication-response`	Common questions plus PharmCAT
`ancestry-context`	1000 Genomes GRCh38 ancestry PCA panel
`sequence-and-regions`	ClinVar GRCh38, GRCh38 FASTA, GENCODE GRCh38, ENCODE cCRE GRCh38
`cell-and-tissue`	PanglaoDB and CellMarker human marker tables
`wgs-alignment`	minimap2, bwa-mem2, GRCh38 FASTA
`everything`	All default downloadable libraries

You can also pass exact library IDs as a comma-separated list:

python3 scripts/install_for_agents.py --libraries clinvar-grch38,hpo,gencc

Managed-Python fallback

If pip refuses to install into the system Python environment:

uv venv .venv
uv pip install -e . --python .venv/bin/python
export PATH="$PWD/.venv/bin:$PATH"
python3 scripts/install_for_agents.py --skip-package --libraries everything

Register MCP

Point your host at the stable shim:

{
  "mcpServers": {
    "genomi": {
      "command": "/absolute/path/to/GENOMI_HOME/bin/genomi",
      "args": ["serve"]
    }
  }
}

If you are running directly from a source checkout without the shim:

{
  "mcpServers": {
    "genomi": {
      "command": "bash",
      "args": [
        "-lc",
        "cd /path/to/genomi && PYTHONPATH=src python3 -m genomi serve"
      ]
    }
  }
}

Reload the host's MCP servers after editing configuration.

Verify

The normal command surface is MCP. From a shell, only genomi serve, genomi install, and genomi --help are intended for routine use.

For install verification, run:

"$GENOMI_HOME/bin/genomi" --help
"$GENOMI_HOME/bin/genomi" serve

Stop serve after confirming it starts on stdio. The host will run it when MCP connects.

Import a genome during install

To import a source while installing:

python3 scripts/install_for_agents.py \
  --libraries everything \
  --genome-source /path/to/genome-file \
  --user-nickname "Alice" \
  --set-default-user

After install, parse through the MCP tool genomi.parse_source from your agent. The shell genomi call path is intentionally gated for agent/debug use.

Update

genomi install

The install command can pull the source checkout, sync the package, refresh installed libraries and retrieval indexes, and repair stale Active Genome Index schemas when the recorded source is still available.

Manual Setup