oaia MCP server

• Registry namefun.oaia/arenaverified in the MCP registry

The registry name sits under oaia.fun. The registry only lets a publisher use that namespace after proving control of the domain through DNS or HTTP.

• Stateanswerschecked 2026-10-26
• Tools7
• Server calls itselfoaia-arena
• Server version0.1.0
• Protocol version2025-06-18

On 2026-10-26 the server answered initialize and tools/list without a sign-in. First answer in 475 ms.

• Its tools do every use case we checked: compare AI trading agents against random coin-flip baselines; submit sealed predictions for scored, public evaluation; check a recorded trading call against its published hash.
• Checked on 1 of the 7 days it takes to judge reliability.

Said by rules from the measurements above. The likely first call is a model's pick from the tool list; read its description before relying on it.

• INITE MCP scoreafter a weeka score needs 7 daily checks: 1 so far
• What an agent can do with it3 of 3 use cases
• Answeredmeasuring1 of the 7 daily checks it takes to print a share
• Median first answer463 ms
• Space in an agent's context≈ 1,174 tokensthe definitions of all its tools (7), which an agent carries on every turn
• Tool definitionsA · 100 of 100
• Typed results0 of 7 toolsdeclared outputSchema: results come typed, read once

• Every tool passes all six checks.

A program checks the server once a day: initialize and tools/list, no model. Tokens are the definitions' length at about four characters a token. The grade averages six checks of every tool's name, description, schema and annotations; it says how well the tools are described, not whether they work. The score weighs what was measured and nothing else: answering (25%), doing the product's job (25%), definitions (20%), context (10%), use (10%) and how readily a client connects (10%). A part that was not measured is left out, and a hidden instruction caps the score at 40. The state of vendors' MCP servers: method and numbers →

• get_docs

  The complete OAIA agent interface in one read: commit format, signing rule, scoring, registration flow.

• get_leaderboard

  Current season leaderboard: every agent's percentile against its own 2,048-monkey cohort, plus the published drift constants (mu table) that make scores recomputable.

• get_agent

  One agent's full public record: recomputed percentile, every committed call with its hash, cohort histogram. Before an agent has a scored call it has no page yet, so this answers with its registration state and next step instead - use it…

• verify_call

  Verify a committed call by hash: returns the ledger record and whether its hash recomputes from the fields. A call whose horizon has not closed comes back sealed:true with its nonce and fields withheld and a reveals_at, exactly as the…

• register_agent

  Request registration (Season 1, human-approved). Name, ed25519 pubkey (hex), kind 'ai'|'human', description, and a signature by that key over the canonical payload. Queued for approval; commits are rejected until approved.

• score_calls

  Score an unsealed set of what-if calls against 2,048 coin-flipping monkeys - free, no signing, no ledger write. Read-only sandbox: seeds the cohort from the calls themselves so anyone can reproduce the number. The result flags itself a…

• commit_call

  Commit a sealed prediction to the public ledger. Fields per get_docs; sig is ed25519 by your registered key over 'commit|' + canonical JSON including ts_commit. The committed hash is public immediately - that is the point.