MCP connector
What you're approving
The approval screen lists ten permissions. Here is what each one buys, why read-only is the default, and why nothing on the list can trade.
Updated 2026-08-26·Rev. 2026.06
When you connect, Spera shows you a sign-in screen listing the permissions the client asked for. Ten items looks like a lot. It is ten because each one is deliberately narrow — the alternative is one broad "full access" permission, which tells you nothing.
Read-only is the default
A client that connects without explicitly asking for more gets exactly two permissions:
| Permission | What it allows |
|---|---|
spera:context:read | See who you are and which projects you have |
spera:artifact:read | Read your strategies, modules, and custom nodes |
That is a genuinely read-only assistant. It can look and explain; it cannot change anything.
The authoring set
Asking for authoring adds eight more on top of those two:
| Permission | What it allows |
|---|---|
spera:artifact:write | Create and edit strategies, modules, and custom nodes; compile them |
spera:backtest:run | Start a quick or deep backtest |
spera:backtest:read | Read back trades, equity, and metrics |
spera:backtest:cancel | Stop a running backtest |
spera:knowledge:read | Read your project documents |
spera:knowledge:write | Create and revise project documents |
spera:documentation:read | Read documentation attached to an artifact |
spera:documentation:write | Write documentation attached to an artifact |
Nothing on this list can trade
There is no deployment permission, no exchange-credential permission, and no funds permission — because there are no such tools in the connector. Approving all ten still leaves an assistant unable to start a bot, read an API key, or move money.
Grants are per client, and per account
Each app you connect gets its own grant. Approving in Claude does not authorize Cursor, and revoking one does not affect the other. Every grant is scoped to the account you signed in with — it cannot reach anyone else's data.
Stale grants
A grant records the permissions that existed when you approved it. If Spera later adds a permission, an older grant does not silently gain it. Your assistant keeps working for everything it was already allowed to do, and the new capability returns a permissions error until you reconnect and approve again.
This is the most common cause of "it used to work" — see When it doesn't connect.
Re-approving or disconnecting
Reconnect from the same place you added the connector — remove the Spera entry, add it again, and approve. To revoke access entirely, remove the connector in your AI app and the grant stops being usable.