omp-tracecat

Tracecat integration for oh-my-pi (omp): the Tracecat MCP server plus five Agent Skills that teach an agent to operate Tracecat safely — workflows, actions, cases, tables, integrations, agent presets, Slack bots, and workspace skill uploads.

Ported from TracecatHQ/tracecat-plugins (Apache-2.0) and repackaged in the portable Agent Plugins 1.0.0 layout omp loads natively: a root plugin.json manifest, a root mcp.json connection, and skills/<name>/SKILL.md packs.

Install

omp plugin marketplace add git@github.com:anhhung04/personal-toolkit.git
omp plugin install tracecat@personal-toolkit

Then open a session. The plugin contributes:

the MCP server tracecat:tracecat (Tracecat Cloud: https://platform.tracecat.com/mcp), whose tools are exposed as mcp__tracecat_tracecat_<tool>;
the skills below, reachable via /skill:<name> or read skill://<name>.

Skills

Skill	Use when
`tracecat-mcp`	Entry point: connecting to Tracecat MCP, workspace discovery, tool routing, safe platform navigation
`tracecat-automation-best-practices`	Workflow DSL/YAML authoring, tables, expressions, run-python, agent presets, graph shape, executions
`tracecat-slackbot-best-practices`	Slack-facing automations: app mentions, interactivity, event subscriptions, thread replies, tone
`tracecat-manage-skills`	Uploading or updating a skill directory through Tracecat MCP (bundled helper script builds the upload payload)
`tracecat-workspace-chat`	Only inside Tracecat's in-product Workspace Chat; its bundled product docs are injected by the Tracecat application image and are absent here

Connect and authenticate

Tracecat Cloud uses OAuth:

/mcp list                              # show tracecat:tracecat
/mcp reauth tracecat:tracecat          # browser login; the credential is stored per omp profile

For a personal access token, or for a self-hosted deployment, add an omp-native override in ~/.omp/agent/mcp.json and hide the bundled cloud entry, then /mcp reload:

{
  "mcpServers": {
    "tracecat": {
      "type": "http",
      "url": "https://platform.tracecat.com/mcp",
      "headers": { "Authorization": "Bearer <your-token>" }
    }
  },
  "disabledServers": ["tracecat:tracecat"]
}

Full details: skill://tracecat-mcp/references/connection.md. Never commit the token or place it in a skill file.

Verify

omp read skill://tracecat-mcp/SKILL.md     # skills discovered
omp plugin list                            # plugin enabled

In-session, /mcp list shows tracecat:tracecat, and a call to mcp__tracecat_tracecat_list_workspaces is the live connectivity check.

Layout

omp-tracecat/
├── plugin.json          # Agent Plugins 1.0.0 manifest (omp agent-plugins provider)
├── mcp.json             # portable MCP connection
├── package.json         # omp runtime package (@anhhung04/omp-tracecat)
└── skills/              # five Agent Skills (SKILL.md + references + scripts)

Upstream sync

TracecatHQ/tracecat-plugins is the authoring home; this repo mirrors it at upstream version 0.5.0. Content is kept verbatim except for the port adjustments:

$skill-name cross-references → omp's /skill:<name> syntax;
Codex skills/*/agents/openai.yaml interface stubs, and the Claude/Codex native manifests (.claude-plugin/, .codex-plugin/, .mcp.json), dropped — omp reads the portable plugin.json + mcp.json instead;
omp connection guidance added to tracecat-mcp (SKILL.md, references/connection.md).

To re-sync: copy plugins/tracecat/skills/ from upstream, re-apply the reference rewrite and drop agents/ subdirectories, then bump version in package.json and plugin.json together.

License

Apache-2.0, retained from TracecatHQ/tracecat-plugins. Tracecat is a trademark of its respective owner.