Every user gets their own scoped, revocable connection. Agents only call the actions you approve — validated, rate-limited, confirmed when risky, and logged every time.
{
"tool": "create_record",
"user": "usr_4h9 · scoped",
"risk": "write",
"status": "allowed"
}
Every user gets their own connection they can revoke anytime — no shared keys, no setup docs. It works from a phone and feels like signing in.
AI tools connect through a standard web sign-in flow — but most mobile apps don't have one, and building it just for AI is a lot of work. mobcp hosts that page for you. Users approve access on a secure screen and land back in the agent, connected. Nothing to build.
// input schema { "type": "object", "properties": { "title": { "type": "string" }, "priority": { "type": "string" } }, "required": ["title"] }
Give an action a name, a JSON schema, a risk level, and a REST mapping. mobcp turns it into a validated, permissioned MCP tool — agents call approved actions like create_record, never your raw API.
Deletes, bulk edits, and purchases can be blocked or held for the user's confirmation. Rate limits stop loops, retries, and runaway cost before an agent runs up your bill.
What the agent called, for which user, whether it succeeded, failed, was blocked, or required confirmation.
Ship "Connect to AI" this sprint — account linking, approved actions, confirmations, and logs included.