Skip to main content
Your agent decides which tools to call, but it helps to know what is available, especially which tools can change your workspace.

​
Read-only tools

​
Tools that change your workspace

These three can modify your workspace. Most MCP clients ask for confirmation before a tool call, keep that on if you want to review changes first.
Nothing in this list can manage connections, datasource credentials, billing, members, or organization settings.

​
SQL behavior

run_sql accepts SELECT, WITH, SHOW, DESCRIBE, and EXPLAIN. Anything that writes is rejected before it reaches your database.
  • Mora appends LIMIT 100 unless your query already sets a limit.
  • The agent receives a 10-row sample by default; it can ask for more, up to the 100-row execution cap.
  • A query that runs longer than 60 seconds keeps executing in the background, and the agent is told to narrow it instead of waiting.
  • Each query is recorded as a run attributed to your user, so MCP activity shows up alongside queries run in the app.

​
Building through Mora’s agent

For multi-step work (“build me a revenue dashboard”), the intended flow is propose, review, then approve:
  1. Your agent calls plan_with_agent, and Mora’s agent researches your data and returns a plan.
  2. Your agent shows you the plan.
  3. You accept or reject, and your agent calls respond_to_plan. Accepting builds it; rejecting sends your feedback back for a revised plan.
ask_agent skips the review step and lets Mora’s agent act directly, useful when you already know what you want. Both share a conversation thread, so follow-up calls continue where the last one left off. Agent runs have a 5-minute budget; if one runs out, the thread id comes back so the work can be continued rather than restarted.
The same thread is visible in Mora’s in-app chat, so you can open the conversation in the browser and see the plan card and everything the agent did.