Connect to an AI agent

The toolkit ships an MCP server, so an AI agent — Claude Code, Codex, VS Code, Claude Desktop, or anything else that speaks MCP — can run the off-target search and guide-design tools directly, instead of guessing at them. Connect it one of two ways.

By URL · nothing to install

rgetoolkit.com runs the server. A download comes back as a link, since a server on our side cannot write to your disk.

claude mcp add --transport http rgetoolkit-remote https://rgetoolkit.com/mcp

codex mcp add rgetoolkit-remote --url https://rgetoolkit.com/mcp

{ "servers": { "rgetoolkit": { "type": "http", "url": "https://rgetoolkit.com/mcp" } } }

In claude.ai or Claude Desktop, add the same URL (https://rgetoolkit.com/mcp) as a custom connector.

As a local package · results on your disk

Runs on your own machine and talks to rgetoolkit.com over HTTP, so a download lands on your disk instead of arriving as a link. Needs uv — uvx fetches and runs the server itself.

{
  "mcpServers": {
    "rgetoolkit": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/pnucolab/rgetoolkit#subdirectory=mcp", "rgetoolkit-mcp"]
    }
  }
}

{
  "servers": {
    "rgetoolkit": {
      "type": "stdio",
      "command": "uvx",
      "args": ["--from", "git+https://github.com/pnucolab/rgetoolkit#subdirectory=mcp", "rgetoolkit-mcp"]
    }
  }
}

[mcp_servers.rgetoolkit]
command = "uvx"
args = ["--from", "git+https://github.com/pnucolab/rgetoolkit#subdirectory=mcp", "rgetoolkit-mcp"]

Point RGETOOLKIT_API_URL at your own deployment to use these tools against it instead of the public site.

Run the toolkit yourself

The whole stack — SvelteKit frontend, FastAPI backend, Celery workers, Redis — ships as a Docker Compose project. It requires Docker with the NVIDIA container toolkit: the search binary JIT-compiles its CUDA kernel, and compose.yml reserves 2 GPUs.

cp .env.example .env      # then fill in SITE_ROOT, GENOME_DIR, SMTP, admin login
docker compose build
docker compose up -d

• Site: http://localhost:13789 — Caddy fronts the frontend and /api/v1/*.
• Admin console: http://localhost:13790 — reference-genome inventory, the request queue, and the notification-email setting. Behind a login; never exposed publicly.

The search binary

Every search tool shells out to cas-offinder-rust (Cas-OFFinder 2), a separate project that is not committed here. Download a build from its releases — or compile your own with cargo build --release — then drop it in:

cp <the-binary> cli/cas-offinder-cli && chmod +x cli/cas-offinder-cli

compose.yml bind-mounts ./cli, so replacing the binary takes effect on the next job with no rebuild. Reference genomes go under GENOME_DIR; see the repository for the full setup.