aimarket-mcp is an ecosystem-level MCP (Model Context Protocol) gateway that centralizes general-purpose web scraping, web search, and Metis verification capabilities behind a single SSRF-hardened entry point, for reuse by Metis, ARGUS, and the entire ecosystem. Its design philosophy is: general-purpose capabilities are written only once, with the security core centralized in one audited location; ecosystem-specific capabilities live in their respective MCP servers.
[Main Capabilities] The gateway provides five hardened tools, among which the three tools web_fetch, web_search, and metis_verify are explicitly mentioned. In addition, two marketplace tools, market_search and market_invoke, are available through the hosted endpoint.
[Transport Modes] Two transports are supported: stdio (entry point is aimarket_mcp/stdio_server.py), for clients such as Glama, Claude Desktop, and Cursor; and Streamable-HTTP (port :9090), for self-hosted deployments (aimarket-mcp-http, Docker Compose). The project is built on the official Model Context Protocol Python SDK (mcp, FastMCP).
[Hosted Endpoint and Trial] If you just want to try out the marketplace features, there is no need to install anything: paste https://modelmarket.dev/mcp into an MCP client to use market_search and market_invoke against the live hub. Each caller gets a small number of free trial calls, and each call comes with a signed receipt, with no wallet, keys, or installation required. Install this package when you need the three tools web_fetch, web_search, and metis_verify, or when you want to point the marketplace tools at your own hub.
[Ecosystem Integration] Metis: set enable_mcp_tools: true and set mcp_ecosystem_presets to [aimarket-web]. ARGUS: add this server to mcpServers in argus.config.json.
[Usage] In stdio mode, the mcpServers entry in Claude Desktop can use uvx to directly pull the published package and run the aimarket-mcp console script, with no need to clone the repository or set a working directory; if the package is already installed, use "command": "aimarket-mcp" with no arguments. Running from a checked-out repository is for development. The self-hosted HTTP mode can be connected to by Cursor / Claude (Streamable HTTP).
[Glama Publishing] The project's listing on Glama is glama.ai/mcp/servers/alexar76/aimarket-mcp, using the same pattern as aimarket-oracle-gateway: glama.json in the repository root + Dockerfile + python -m aimarket_mcp.stdio_server. Note not to use the auto-generated debian:trixie-slim + uv sync + mcp-proxy template (that path has HTTP/ENOENT issues). Glama ignores the repository Dockerfile, so the build steps must be set in admin/dockerfile: CMD is [".venv/bin/python", "-m", "aimarket_mcp.stdio_server"], pin main or the tag glama-build (do not pin SHA).
[Repository Notes] This is a read-only mirror; aimarket-mcp is published from the authoritative AI-Factory monorepo, pull requests are not accepted, and any commits pushed here will be overwritten by scripts/mirror_satellites.sh on the next sync. If you find a bug or have a request, please open an issue.
[On-Chain Hosting Flow] The example mentioning "buying a GAIA Open-Meteo reading" is the simple path (trial/sandbox). If you need to prove that an external wallet completed payment on-chain, that is a different hosting flow (openChannel → DebitAuthorization → hub debit/settle); see docs/onchain-journal.md §3l–§3m and gaia/docs/LIVE-RELAYS.md for details.
[Other Information] The project provides multilingual documentation entry points (English, Русский, Español, Français, 中文, Glossary), and includes a code of conduct and security policy. Regarding specific pricing, free quota details, etc., the materials only mention "a small number of free trial calls"; the rest is not yet provided.
