What the work actually is
RL Studio needs real software running inside reproducible sandboxes so agents can be trained and evaluated against it. Your job is to make that happen app by app: stand up an MCP server in Python using FastMCP, expose the application's meaningful actions as tools, wire the app into a container with its services and dependencies, and implement populate/snapshot hooks so environment state can be seeded and restored deterministically. A typical unit of work is one app, taken from "here is the tool we want covered" to "it passes platform validation and deploys."
Most of the difficulty is not novel algorithms — it is the unglamorous middle: a desktop app that expects a display server, an API that rate-limits, a database that must be reset between rollouts, file permissions that differ between your machine and the built image. Debugging build, runtime, and permission failures is a large share of the day.
What the screen looks for
- Real Python depth, not scripting familiarity — async, packaging, error handling in long-running services
- Concrete Docker and Linux experience: writing Dockerfiles, diagnosing why something works locally and fails in the image, understanding processes, users, and mounts
- Evidence you have integrated third-party or closed software you did not control, and how you handled undocumented behavior
- Judgment about determinism and testability — whether you think about state reset, idempotency, and reproducibility unprompted
- FastMCP or MCP exposure is a plus, not a gate; the screen will probe whether you understand the tool-server pattern generally
Logistics
Fully remote and largely asynchronous, contract-based, with work assigned as app-scoped tasks. Observed pay for this listing is $40–50/hr, varying with experience and assessed skill; rates are set by the platform and are not guaranteed. Expect a technical screen plus a practical exercise, and expect to state honestly how many hours per week you can commit — inconsistent throughput is a common reason contracts end early.