For teams who ship agents
A hosted sandbox for the agent you are building.
Mockr hosts the world around it: stateful fake tools over HTTP and MCP, scripted teammate agents, and stubbed connections to remote agents and models.
Three surfaces
-
Tools
Stateful fake tools over HTTP and MCP. A tool records a session, follows a script, or serves a fixture. State carries from one call to the next, and the same fixture returns the same result on a laptop, in an eval, and in CI.
-
Teammates
Scripted stand-ins for the other agents in the group. Each has a role, a way of answering, and the failures the scenario asks it to produce. The agent under test talks to a group. The team writes the teammates.
-
Connections
Stubbed routes to remote agents and model providers. A connection returns a recorded completion, a scripted reply, or a fixture pinned for the run. Point it at the real service when the run should leave the sandbox.
How a team uses it
Point the agent you are building at a Mockr sandbox. Walk a scenario locally, score it in an eval, and run the same fixtures in CI. Record a behavior you want to keep, and swap a stand-in for the live connection when the run should call the real service.