Om MCP Overview

Connect Codex, Claude Code, and other MCP-compatible clients to the hosted Om MCP with your Om API key.

Om MCP is the hosted MCP interface for Om. Add it to Codex, Claude Code, or any MCP-compatible client, authenticate with your Om API key, and use Om tools directly from the same session.

Setup in four steps

1

Create an Om API key

Generate an API key from your Om dashboard and keep it in an environment variable.

export OM_API_KEY="omtx_..."
2

Add the hosted Om MCP

Point your MCP client at the hosted Om MCP endpoint.

3

Choose the auth mode your client supports

Codex can use `Authorization: Bearer $OM_API_KEY`. Claude Code and other clients with custom-header support can send `x-api-key: $OM_API_KEY`.

4

Restart and verify

Ask for `om_status`, `pricing_get`, or `datasets_catalog` to confirm the server is attached.

Hosted server

SettingValueNotes
MCP URLhttps://agents.omtx.ai/mcpHosted Om MCP endpoint. Verified live in this workstream.
Transportstreamable HTTPMCP clients should connect over HTTPS to the /mcp route.
Codex authAuthorization: Bearer $OM_API_KEYRecommended for Codex because the current remote MCP setup supports bearer-token auth directly.
Custom-header authx-api-key: $OM_API_KEYRecommended for Claude Code and other clients that support custom headers.

Add Om MCP to your client

codex mcp addBash
codex mcp add omtx --url https://agents.omtx.ai/mcp --bearer-token-env-var OM_API_KEY
~/.codex/config.tomltoml
[mcp_servers.omtx]
url = "https://agents.omtx.ai/mcp"
bearer_token_env_var = "OM_API_KEY"

Diligence

Search, gather, crawl, synthesize, and run deep diligence from the same session.

  • Fast target triage
  • Claim-backed research workflows
  • Citation-oriented outputs

Data Access

Find available proteins and retrieve binder and non-binder dataset exports for a protein.

  • Protein discovery
  • Signed binder and non-binder exports
  • Model-ready dataset pulls

Jobs + Artifacts

Launch work, wait on completion, export results, and retrieve artifacts without leaving the session.

  • Polling and wait helpers
  • Artifact upload and finalize
  • Signed download URLs

Hub

Launch supported model workflows and inspect execution state directly.

  • Submit runs
  • Track job progress
  • Pull outputs into downstream work

System

Check health, pricing, profile, and credits from the same client session.

  • om_status
  • pricing_get
  • credits_get
  • users_profile

Good first prompts

  • Run om_status to verify the MCP is attached.
  • Run pricing_get to inspect the current pricing manifest.
  • Ask Om MCP which proteins and datasets are available in Data Access.
  • Run diligence_search or diligence_gather on a target you already care about.

Typical MCP workflow

1

Ask the question

Start with a target, protein UUID, job goal, or workflow question inside your MCP client.

2

Use Om tools directly

Run Diligence, Data Access, Jobs, Artifacts, or Hub tools without building request glue first.

3

Keep the result in-session

Let the next step use the returned evidence, dataset export URLs, job IDs, or artifact URLs immediately.

4

Move into the API when needed

Use the direct API only when you need orchestration beyond the MCP session.

Security and approval

Om MCP can access your Om account with the same API key you use against the API directly. Keep manual tool approval on in your MCP client, use a dedicated API key per machine or environment, and do not commit that key to source control.

QuickstartAvailable ToolsExample PromptsConfiguration Options