Om MCP Available Tools
The hosted Om MCP exposes Diligence, Data Access, Jobs, Artifacts, Hub, and System tools through one remote MCP server.
Om MCP gives MCP-compatible clients a direct tool surface over Om. You do not need to memorize every tool name to use it well, but it helps to know how the tool families are organized and what kinds of questions each family can answer.
Tool families
| Family | What it does | Best for | Representative tools |
|---|---|---|---|
| Diligence | Searches, gathers, crawls, synthesizes, and deepens target research. | Target triage, mechanism reviews, claim-backed research briefs | diligence_search, diligence_gather, diligence_crawl, diligence_deep_diligence, diligence_synthesize_report, rag_search |
| Data Access | Shows available proteins and returns binder and non-binder dataset export URLs for a protein. | ML workflows, data review, candidate prioritization | datasets_catalog, datasets_generated_protein_uuids, binders_get_shards, binders_urls |
| Jobs | Tracks asynchronous work, waits on completion, exports results, and retrieves artifacts. | Operational workflows and job orchestration | jobs_history, jobs_status, jobs_wait, jobs_export_json, jobs_get_deep_diligence_result, jobs_get_artifact_url |
| Artifacts | Uploads files, creates signed upload URLs, finalizes uploads, and resolves artifact metadata. | Hub inputs, notebooks, downstream file handoff | artifacts_upload_bytes, artifacts_create_upload_url, artifacts_finalize_upload, artifacts_get |
| Hub | Launches active model workflows through the Om Hub tool surface. | Model execution from inside Codex or Claude Code | hub_submit and active hub_<model> launch helpers |
| System | Checks health, pricing, credits, and account profile state. | Verifying setup and inspecting account context | om_status, pricing_get, credits_get, users_profile |
Natural-language asks that map well to Om MCP
Research and diligence
Ask natural questions about targets, mechanisms, competitors, or pathways.
- “Search KRAS and summarize the therapeutic strategy.”
- “Run deep diligence on KEAP1 resistance.”
- “Gather the strongest evidence for EGFR resistance mechanisms.”
Protein and dataset retrieval
Ask for available proteins or for binder and non-binder datasets for a specific protein UUID.
- “What proteins can I access in Data Access?”
- “Give me molecular binders for this protein UUID.”
- “Show me non-binders too so I can compare the dataset.”
Jobs and artifacts
Ask the client to launch work, wait on it, and return the result or the artifact URLs.
- “Launch this workflow and wait until it completes.”
- “Export the result as JSON.”
- “Get the signed artifact URL for the output.”
Hub execution
Launch active Hub models from the same session where you did the research.
- “Launch a Hub workflow for this target.”
- “Use the uploaded artifact in the next run.”
- “Track the job until it finishes.”
Notes on tool names
- Most users should start with natural-language asks and let the MCP client choose the tool.
- Some tool names remain transport-oriented because they mirror the canonical Om API surface.
- Customer-facing docs should describe outcomes in plain language, even when the underlying tool returns signed dataset export URLs or job IDs.
- If you need strict orchestration or exact payload control, move from MCP into the direct API.