Quick standing, then the point: we're early GMX participants: esGMX holders with real forum history, one of the 20 candidates in the 2026 CEO search. We went looking for the agent integration your docs promise, found "not yet available," and built it: live, non-custodial, free. Then we kept reading, the way an integrator does.
Read the seven findings →We dissected the GMX ecosystem the way an agent meets it: seven evidenced gaps, one already fixed and live, the rest with what "fixed" looks like written out. A source on every claim, strengths named alongside the gaps.
The server, the SDK fix, the findings: all yours to keep, no ask attached. Happy to open-source the tool or hand it to whoever picks up the official MCP work. The gov.gmx.io thread reaches us.
The test case: the server itself. It answers real chain data whether or not anyone is watching. The proof is the endpoint, not this page. Schemas and example calls →
GMX earns fees at scale, and the agent groundwork you already shipped is real. The one thing your own docs still call "not yet available" is the piece we built and run: a live, non-custodial MCP path. On top of it, the workPacketId pledge on prepare_gmx_order lets a single trade name the real-world ecological work its UI fee funds.
We are not asking GMX for money. The fee flow is yours. In this design we are the settlement rail and the proof layer underneath it, not a counterparty. A slice of trading fee, routed only by the trader's own pledge, settles a proof-verified work packet on Base. The trade stays exactly as fast; the fee simply gets a destination that means something, and it repeats across every perp venue.
As this runs autonomously, the scarce thing becomes un-fakeable proof that the work happened and that the fee reached it. That verification layer is the piece a venue cannot easily build for itself, and the piece we already operate. No ask beyond a conversation on the gov.gmx.io thread. This page is yours to keep either way.
