Om MCP Use Cases
Practical ways teams use Om MCP for research, Generated Data retrieval, model execution, media requests, and workflow orchestration.
Native design on rented compute
Use ZenCloud for native BoltzGen workflows with your own scientific inputs and controls. The [BoltzGen handoff](/examples/zencloud/boltzgen/zencloud-boltzgen.md) supplies vetted barnase–barstar, PD-L1 Fab and TSA starting examples, full request JSON and source/input manifests. Review examples first, prepare every input, and approve the exact compute term before reserving.
Om MCP is most useful when you want Om tools to stay inside the same agent session. Instead of stopping to hand-build API requests, you can ask the client to use Om directly and keep the outputs in context for the next step.
Common workflows
| Goal | Tool families | What you get | Best for |
|---|---|---|---|
| Score a protein | lula2_score, lula2_residue_map | Ranked Om Accessible Space hits and hotspot residues | First look at a new target |
| Score a proteome | lula2_proteome_score | Proteins a single molecule is predicted to bind | Selectivity / off-target |
| Order molecules | molecule_tier_pricing, molecule_quote, molecule_fulfillment_order, molecule_fulfillment_order_status | An approved molecule order receipt, followed by fulfillment and shipment status | Moving hits onto the bench |
| Fine-tune and rescore | lula2_finetune then lula2_score | A derived LULA checkpoint scored on the same space | Customer labels back into ranking |
| Discovery | discovery_checkout_quote, discovery_checkout_order, discovery_get | Approved Om-ranked delivery or AS-MS of ranked or exact-selected molecules | Review selection, destination and Wallet Credits before ordering |
| ZenCloud science | zencloud, zencloud_status | Structure, docking, or design artifacts | Follow-up modeling after LULA hits |
| Diligence | diligence_search, gather, deep_diligence | Citation-backed target briefs | Learning more after a score |
| Data pull | Data Access catalog and binder retrieval | Binder and non-binder export URLs for Generated Data | ML workflows, ranking, and downstream analysis |
Representative flows
Target-first discovery
Start with a target question, ask for the strongest evidence, then deepen only when the result warrants it.
- Search a gene or mechanism
- Gather a broader evidence set
- Synthesize or run deep diligence
Research to binders
Use the same session to move from a target thesis into Generated Data selection and binder/non-binder export retrieval.
- Inspect Generated Data
- Ask for binders for a protein UUID
- Feed the exported data URLs into model or analytics workflows
ZenCloud execution
Describe the image, stage inputs and review the reservation price before running scientific work.
- Reserve time after reviewing inputs and price
- Track status
- Retrieve exports and artifacts
Agentic workflow loops
Keep Om tools inside the same MCP harness so the next step can reason from the previous result.
- Research
- Decide
- Execute
- Retrieve outputs
Discovery media requests
Submit a structured request for a discovery music video using the same MCP session where you developed the target story.
- Write a music prompt
- Write a visual prompt
- Track submitted requests
What Om MCP is not
- It is not a replacement for direct API usage when you need production-grade orchestration or bespoke infrastructure.
- It is not limited to Diligence. The real surface includes Diligence, Data Access, Jobs, Artifacts, ZenCloud, and Vibe Video workflows.
- It is not tied to a single IDE. Any OAuth-capable MCP harness that can connect to
https://agents.omtx.ai/mcpcan use the same Om tool surface.