Senior Software Engineer (MCP Environments)

What you'll do

This is the agentic version of micro1's software engineering environment work. You still build reinforcement learning environments around real engineering problems (a bug fix, a feature, a refactor, a performance fix), but the model being tested cannot solve them from the code alone. It has to discover and reason over information held behind real Model Context Protocol (MCP) servers, the tool interfaces that let an AI agent query issue trackers, documentation, databases and similar systems.

That changes the design problem. A good task makes tool use necessary rather than decorative, and the scoring has to measure two things at once: whether the agent used its MCP tools sensibly and whether the final code is correct. The ad asks for reproducible environments, deterministic verification and golden reference solutions, so you are writing both the reference answer and the checks that decide whether an attempt passes.

Who fits

• Proficiency in C++, Python, Java, Go, TypeScript or Rust
• Deep algorithms, data structures and performance tuning knowledge
• Experience debugging complex issues and delivering maintainable fixes
• A solid background in feature development and codebase refactoring
• Proven optimisation for performance and scalability
• Strong written and verbal communication, and experience in collaborative, ideally remote, teams

Preferred: work on large distributed codebases, rigorous code review, and some familiarity with modern AI or ML systems (a plus, not required). The ad does not ask for prior MCP experience, but engineers who have already wired tools into an agent will find the task design much easier.

What it pays

$80–120/hour is the published band, paid on output: per task that meets the project specifications, with a weekly minimum number of submissions that the ad does not quantify. Expect around 15 hours a week. Deterministic verification for tool-using agents is fiddly, so early tasks will take longer than the band suggests.

Process

Screening questions, an AI interview of about 30 minutes, a technical assessment (marked tentative) and a hiring manager review. micro1 says roles usually fill within 48 hours and first tasks start 24–48 hours after onboarding.

How it differs from micro1's other Senior Software Engineer listings

The MCP tool-use layer is what sets this apart. micro1's RL environment posting at $100–150/hour and its $50–100 twin build plain code environments without the agent-tooling requirement. The Coding Tasks and Verifiers listing focuses on verifiers at $30–100, and the $245–280 listing is code evaluation gated on big-tech experience.

Worth knowing

Good:

• Current, agent-focused work that builds skills worth having
• 100 openings, flagged high demand
• About 15 hours a week
• The hiring steps are published up front
• No prior AI experience required

Less good:

• Per-task pay with an unquantified weekly minimum; rejected tasks are unpaid
• Two things to verify per task (tool use and code) means more design effort per payout
• A technical assessment may be added
• Contractor engagement: no benefits, your own tax to manage

About this listing

Posted by micro1 on its own jobs portal about two months before we read it on 24 September 2026. The $80–120/hour band, the 100 openings, the 15-hour weekly guide, the process steps and the per-task terms are its published text. Location is Remote with no country list, and the customer is not named. Payment schedule and method are not published on the listing. See micro1.