Reverie R1

Open-source tooling to train classifiers and fine-tune adapters on top of foundation models for life sciences.

Evals for clinical genetics

An evaluation suite built by experts in clinical genetics to test whether AI models can perform clinical workflows.

Agentic tools forbinder design, structure prediction, molecular dynamics, docking, protein design, and MSA generation

Computational biology workflows, exposed as MCP tools. One server, typed I/O, pay per run.

agent session · reverie-mcp
User
Design 8 binders for ubiquitin around residues 48, 49, 68.
Agent · thinking
Known target. I'll call reverie.design_binder.
Tool call
reverie.design_binder({ target_pdb: "1ubq", hotspot_residues: ["A48", "A49", "A68"], num_designs: 8 })
RFdiffusion✓ 8 backbones2m 04s
ProteinMPNN✓ sequences18s
AlphaFold2✓ refold + pLDDT3m 47s
filter✓ 5 of 8 pass< 1s
5 ranked binders · pLDDT 84–91 · $4.80
Agent · reply
5 binders passed. Top design pLDDT 91, clusters at the hotspots. Run OpenMM relaxation on the top 3?
Structure & affinity

Structure and binding affinity in one call.

Boltz-2 GPU · ~30–90s
Metered
per run
Binder design

RFdiffusion → MPNN → AF2 filter as one job.

Chained pipeline GPU · ~5–10m
Metered
per run
Small molecule docking

Blind docking with PoseBusters QC.

DiffDock GPU · ~1–3m
Metered
per run
Molecular dynamics

Relaxation, minimization, equilibration.

OpenMM GPU · ≤ 10ns
Metered
per GPU-min
MSA generation

mmseqs2 against UniRef + environmental DBs.

ColabFold CPU · ~30–120s
Free
on free tier
Protein embeddings

Per-residue or per-sequence embeddings.

ESM-2 CPU · < 1s
Free
on free tier

Connect in one line.

Add the server to your agent config — Claude Desktop, Cursor, Claude Code, or any MCP-compatible runtime.

{
  "mcpServers": {
    "reverie": {
      "url": "https://mcp.reverieresearch.com"
    }
  }
}
Onboarding beta partners