Agents and keys
One address for Claude, Cursor or any agent, with everything it does logged like a colleague's work.
An agent gets a key, the same as a person. A key fits one dataset at a time, at one of three levels: add-only (it can post rows and read nothing), read, or read and write. Add masked and names, emails and the like are replaced by stable stand-ins. Keys can expire, and every one can be revoked in a click on the Agents screen.
Connecting Claude Code
claude mcp add --transport http overscore https://app.overscore.tech/mcp \ --header "Authorization: Bearer ovs_…"
The agent gets eight tools: the guide (read first), notes, table search, describe, query, insert, write a note, list tables. Its standing instruction is to use signed-off tables first and models second, and to read the business context before it reads the data.
What's logged
Every query, every insert, every note, under the agent's own name, with the tokens and cost. The Agents screen shows what each one did and what it spent. An agent under your account is never anonymous.
Limits per key
- 1,000 rows returned per query, 30 seconds per query.
- 120 requests a minute, 20,000 a day.
- All adjustable per key when a job needs more.