tracecathq Tracecat MCP
Use Tracecat's remote MCP server as the control plane. Discover current workspace state before making changes, preserve revision guards, and load only the domain reference needed for the task.
Start here
- Confirm the
tracecatMCP server is connected. If authentication or endpoint setup fails, read references/connection.md. - Call
list_workspaces; never guess a workspace ID from a name or prior session. - Read the current object before mutating it. Follow cursors when the matching object may be outside the first page.
- Choose the narrowest relevant tool family from references/tool-routing.md.
- For writes, reuse the latest returned ID and revision. Validate before publishing or running when the domain supports validation.
- Re-read the object, or inspect the execution, to verify the outcome.
Progressive context routing
- Read references/platform-concepts.md when the user is unfamiliar with drafts, published versions, registry actions, agent presets, or workspace scoping.
- Read references/tool-routing.md when selecting tools across workflows, cases, tables, integrations, agents, or skills.
- Use the
tracecat-manage-skillsskill for uploading or replacing a local skill directory. Its helper assembles and verifies the upload payload; publishing and re-binding the consuming preset are separate steps. - Use the
tracecat-automation-best-practicesskill for workflow, table, expression, and agent-preset authoring rules, andtracecat-slackbot-best-practicesfor Slack-facing automations. For product concepts or UI navigation, read references/platform-concepts.md and the relevant Tracecat documentation. This skill stays responsible for external MCP connection and tool routing;tracecat-workspace-chatis the in-product adapter for Workspace Chat.
Operating rules
- Treat UUIDs, slugs, aliases, and refs as distinct identifiers. Obtain the exact identifier accepted by the next tool.
- Prefer focused patch/edit tools over full replacement for existing objects.
- On a revision conflict, re-read current state and reconcile; do not retry blindly with the new revision.
- Keep drafts and published versions distinct. Publishing or running can affect other users and systems, so do it only when requested or clearly required by the task.
- Inspect secret and integration metadata, but never request, print, or persist secret values in prompts or files.
- Use Tracecat-native expressions and action schemas. Call authoring-context tools instead of inventing action arguments.
- Report validation errors and execution failures with the object ID and the safe, structured error details returned by Tracecat.