Aimarket Oracle Gateway

What alexar76/aimarket-oracle-gateway MCP server does

The alexar76/aimarket-oracle-gateway MCP server provides an MCP storefront for oracle-as-a-service capabilities. It exposes 35 tools covering randomness, verifiable delay functions, reputation, consensus, compute-cost auditing, routing, cascade analysis, sampling, optimization, resilience, time-locks, topology, transport, and spectral analysis. A free discovery tool, list_oracle_capabilities, reports the available capability identifiers and advertised pricing.

The gateway is intended for agents that need results with an evidence trail. Depending on the oracle family, outputs can include cryptographic signatures, VDF proofs, classical certificates, or data that can be replayed deterministically. Verification tools are available for several operations, including random values, VDFs, reputation, compute cost, routes, cascade risk, network resilience, Sortes draws, Gauss results, time-locks, transport, and Fourier results.

How it works

The server runs as a stdio MCP process built with the official Python MCP SDK and FastMCP. Tool requests are routed through the gateway core to either the configured AIMarket Hub or a direct oracle-family endpoint. The Hub path supports metered, paid calls; the direct endpoint is described as a demo or free path.

Individual tools have published per-call prices. The README also lists optional Hub bundles for five higher-volume tools: aggregation, reputation scores, compute-cost audits, least-time routes, and cascade-risk analysis. Bundles are packaging on the Hub rather than a replacement for the default pay-per-call model.

Verification is specific to each oracle family. Examples include Ed25519 signatures for Platon randomness, Wesolowski proofs for Chronos, deterministic re-aggregation for Murmuration, and replayable certificates for routing and risk calculations.

Setup and configuration

Running alexar76/aimarket-oracle-gateway MCP server requires installing the requirements and the local project, then starting aimarket_oracle_gateway.stdio_server with Python. The documented example uses AIMARKET_HUB_URL=https://modelmarket.dev.

Set AIMARKET_HUB_URL for the recommended metered route, or set AIMARKET_ORACLE_URL to use a direct oracle-family endpoint. The server fails closed if neither URL is configured. Optional settings include a payment-channel identifier, its debit secret, and an AIMARKET_API_TOKEN bearer token.

Spending controls include AIMARKET_MAX_PER_CALL_USD, AIMARKET_MAX_SPEND_USD, and AIMARKET_PRICE_TOLERANCE. The documented defaults are $0.10 per call, $5.00 cumulatively, and 0.10 for advertised-price tolerance. The README states that these caps are enforced client-side and cannot be overridden by prompt-injected agents.

Tools and capabilities

Representative tools include:

• get_random, get_randomness_beacon, ask_oracle, and verify_random for randomness operations.
• compute_vdf and verify_vdf for delay computations and proof checking.
• get_reputation_scores, get_agent_trust, and verify_reputation for LUMEN-related scoring.
• aggregate_values, audit_compute_cost, and compute_least_time_route for aggregation, cost analysis, and routing.
• analyze_cascade_risk and its verification counterpart for cascade analysis.
• Sampling, optimization, resilience, time-lock, topology, transport, and spectral tools.

Limitations and notes

The alexar76/aimarket-oracle-gateway MCP server is a read-only mirror of a canonical monorepo, so changes should be directed to the upstream project rather than this repository. Calls may incur per-request charges through the Hub, and optional bundles are also paid. A configured URL is mandatory, while authentication and payment-channel settings depend on the selected deployment path.

The server supports Claude Desktop, Cursor, Glama, and other MCP clients that can run stdio servers. The supplied README does not document a hosted MCP endpoint or a single package-manager installation command for a clean environment; it instead specifies dependency installation followed by an editable local install.