MCP Expert

What you'll do

You build reinforcement learning environments that measure two things at once: whether an AI agent can solve a real software engineering problem, and whether it can use Model Context Protocol tools well enough to get there. Tasks look like ordinary engineering work (fix a bug, implement a feature, refactor, optimise performance), but the agent has to discover and reason over information served by real MCP servers to do them.

Your deliverable is the environment itself: reproducible setup, deterministic verification, and a golden reference solution. The verification has to be deterministic, since an environment that grades an agent inconsistently is worse than no environment. That makes the engineering discipline closer to writing a test harness than to writing a task.

The engagement is remote contractor work at roughly 15 hours a week, and you choose your own hours and days, weekends included. No prior AI experience is required.

What you need

Listed as required, not preferred:

• Proficiency in C++, Python, Java, Go, TypeScript or Rust
• Deep understanding of algorithms, data structures and performance tuning
• Demonstrated experience debugging complex software issues and delivering maintainable solutions
• A strong background in feature development and codebase refactoring
• Proven ability to optimise software for performance and scalability
• Exceptional written and verbal communication, with close attention to detail
• A track record in collaborative, cross-functional teams, ideally remote

Notably absent: any requirement for a public open-source profile, a degree, or a particular employer. If you have working MCP experience without either of the first two, this is a listing that will take you.

Nice to have

• Previous work on large-scale, distributed codebases
• Familiarity with modern AI or machine learning systems, stated as a plus, not a requirement
• A background in rigorous code review and in shaping software best practice

What it pays

$60–120/hour, then qualified: pay is output-based, per task that meets the project specification, not per hour worked. The time a task takes is stated to vary with the expert's experience and workflow, so the band describes what a fast contributor reaches rather than a rate you are owed for time spent.

A minimum weekly submission requirement also applies, and the number is missing from the ad: it reads "Experts must submit a minimum of tasks per week". Ask what it is before you accept.

How you get hired

Four stages: apply and answer the screening questions, complete an AI interview of about 30 minutes, sit a technical assessment (marked tentative), and pass a hiring manager review. None of that time is paid.

Roles are stated to fill typically within 48 hours, with first tasks 24 to 48 hours after onboarding.

Worth knowing

Good:

• No degree, no named employer, no open-source profile required, unusual for a role at this rate
• MCP is new enough that practical experience with it is scarce, which gives you an edge on a listing with 100 openings
• You choose your hours and days, weekends included
• The work is real systems engineering: reproducible environments and deterministic verification
• The requirements are stated as requirements, so you can judge your fit honestly rather than guess what "preferred" means

Less good:

• Output-based pay against an advertised hourly band, and building a deterministic environment is slow, fiddly work
• A weekly submission minimum applies, but its number is missing from the listing
• A $60-wide band with no published basis for where in it you land
• Four screening stages, including an unpaid technical assessment
• Contractor engagement: no benefits, no notice period, your own tax to manage
• Expected to start 24–48 hours after onboarding
• Neither the customer nor the project length is named

About this listing

Posted by micro1 on its own jobs portal, which showed it as posted around 23 August 2026; we confirmed it open on 7 September 2026. The $60–120/hour band, the required-skills list and the four-stage process are its own published text for this listing.

No country restriction is named on the ad itself, so eligibility is left open. We could not verify the per-task rate or the weekly submission minimum; neither is published. Screening and payment run at the platform level. See micro1.