refua-mcp 0.5.2

MCP server for Refua Boltz2 folding/affinity and BoltzGen design workflows.

Refua MCP Server

MCP server exposing strict, typed Refua tools for Boltz2 folding/affinity and BoltzGen design workflows.

Install

pip install refua[cuda] # remove [cuda] if you don't need gpu support
pip install refua-mcp

ADMET predictions are optional; install refua[admet] to enable them:

pip install refua[admet]

Boltz2 and BoltzGen require model/molecule assets. If you don't have them, refua can download them for you automatically:

python -c "from refua import download_assets; download_assets()"

• Boltz2: uses ~/.boltz by default. Override via tool boltz.cache_dir if needed.
• BoltzGen: uses the bundled HF artifact by default. Override via tool boltzgen.mol_dir if needed.

MCP Clients

Claude Code

Add the server to your Claude Code MCP config (macOS: ~/Library/Application Support/Claude/claude_code_config.json, Linux: ~/.config/claude/claude_code_config.json). This uses the default assets (~/.boltz for Boltz2 and the bundled BoltzGen artifact). Merge with any existing mcpServers entries:

{
  "mcpServers": {
    "refua-mcp": {
      "command": "python3",
      "args": ["-m", "refua_mcp.server"]
    }
  }
}

Codex

Register the server with the Codex CLI (uses default asset locations):

codex mcp add refua-mcp -- python3 -m refua_mcp.server

List configured servers with:

codex mcp list

If the server is slow to boot (for example on first import of heavy ML deps), raise the startup timeout in your Codex config.toml:

[mcp_servers.refua-mcp]
startup_timeout_sec = 30

Tools

• refua_validate_spec: validate and normalize a request without running folding/affinity.
• refua_fold: run fold/design workflows with typed entities and constraints.
• refua_affinity: run affinity-only predictions.
• refua_antibody_design: focused antibody entrypoint (antibody + optional context_entities).
• refua_job: check status/results for background jobs.
• refua_admet_profile (optional): run model-based ADMET predictions for SMILES strings (requires refua[admet]).

All major tools expose strict JSON schemas, including discriminated entity unions by type and typed output schemas.

Output-format notes:

• structure_output_format: canonical values are cif or bcif; mmcif is accepted as a compatibility alias for cif.
• feature_output_format: use torch or npz when writing feature files. If json is provided with a file output request, the server normalizes it to npz and reports a warning.
• If feature_output_path ends with .json, the server normalizes it to .npz for feature file export.
• Fold/design responses include a warnings list when compatibility normalization is applied.

Input-compatibility notes:

• For compatibility with some MCP clients, entities, constraints, antibody, and context_entities also accept JSON-encoded strings and will be normalized server-side.

Execution-policy notes:

• refua_fold and refua_antibody_design block exploratory names by default (for example sanity_*, *_probe, schema_*, smoke_*) to avoid costly preflight runs.
• If an exploratory run is intentionally needed, set allow_exploratory_run=true.

Resources And Templates

• refua://recipes/index (resource): lists canonical recipe names.
• refua://recipes/{recipe_name} (resource template): returns concrete tool/args examples.

Recipe names:

• fold_protein_ligand
• affinity_only
• antibody_design

Workflow

Recommended call sequence:

1. Read refua://recipes/index and optionally a recipe template.
2. Call refua_validate_spec to catch schema/logic issues before expensive runs (deep_validate=true for asset-backed construction checks).
3. Execute refua_fold, refua_affinity, or refua_antibody_design.
4. For long runs, set async_mode=true and poll refua_job.

Examples

Fold protein + ligand with optional affinity:

{
  "tool": "refua_fold",
  "args": {
    "name": "protein_ligand",
    "entities": [
      {"type": "protein", "id": "A", "sequence": "MKTAYIAKQRQISFVKSHFSRQLEERLGLIEVQ"},
      {"type": "ligand", "id": "lig", "smiles": "CCO"}
    ],
    "constraints": [
      {"type": "pocket", "binder": "lig", "contacts": [["A", 5], ["A", 8]]}
    ],
    "affinity": {"binder": "lig"},
    "admet": true
  }
}

Affinity-only:

{
  "tool": "refua_affinity",
  "args": {
    "name": "protein_ligand_affinity",
    "entities": [
      {"type": "protein", "id": "A", "sequence": "MKTAYIAKQRQISFVKSHFSRQLEERLGLIEVQ"},
      {"type": "ligand", "id": "lig", "smiles": "CCO"}
    ],
    "binder": "lig"
  }
}

Antibody-focused design:

{
  "tool": "refua_antibody_design",
  "args": {
    "name": "ab_design",
    "antibody": {
      "type": "antibody",
      "ids": ["H", "L"],
      "heavy_cdr_lengths": [12, 10, 14],
      "light_cdr_lengths": [10, 9, 9]
    },
    "context_entities": [
      {"type": "protein", "id": "A", "sequence": "MKTAYIAKQRQISFVKSHFSRQLEERLGLIEVQ"}
    ]
  }
}

Validate only (no run):

{
  "tool": "refua_validate_spec",
  "args": {
    "action": "fold",
    "entities": [
      {"type": "protein", "id": "A", "sequence": "MKTAYIAKQRQISFVKSHFSRQLEERLGLIEVQ"},
      {"type": "ligand", "id": "lig", "smiles": "CCO"}
    ]
  }
}

ADMET predictions:

{
  "tool": "refua_admet_profile",
  "args": {
    "smiles": "CCO",
    "include_scoring": true
  }
}

Note: DNA/RNA entities are supported for Boltz2 folding only (BoltzGen does not accept DNA/RNA entities).

Long-Running Jobs

For runs that exceed the tool-call timeout, set async_mode=true and poll sparingly (for example every 30-120 seconds) or follow recommended_poll_seconds from refua_job.

{
  "tool": "refua_fold",
  "args": {
    "async_mode": true,
    "entities": [...]
  }
}

Then poll:

{
  "tool": "refua_job",
  "args": {
    "job_id": "..."
  }
}

For queued/running jobs, the response includes recommended_poll_seconds plus queue and estimate metadata (queue_position, queue_depth, average_runtime_seconds, estimated_start_seconds, estimated_remaining_seconds). Set include_result=true once complete to fetch results.

Long-poll support:

{
  "tool": "refua_job",
  "args": {
    "job_id": "...",
    "wait_for_terminal_seconds": 300,
    "include_result": true
  }
}

Experimental MCP Tasks

This server enables MCP experimental task support (tasks/get, tasks/result, tasks/list, tasks/cancel) and advertises task execution support for refua_fold, refua_affinity, refua_antibody_design, and refua_admet_profile.

If your client supports task-augmented tool calls, prefer tasks for long-running operations. Otherwise, continue with async_mode=true + refua_job.