Sugra API MCP is the official Model Context Protocol (MCP) server for Sugra API, positioned as the connector between LLM agents and world data. Through a single MCP server, it gives any AI agent access to more than 1,600 data endpoints from over 160 first-party sources, covering 36 data domains, with data ranging from markets, economics, companies, government, news, climate, maritime, and entity screening.
In terms of main capabilities, the server provides a bundled endpoint catalog, and discovery-type operations (search, describe, toolsets) run locally without network calls; only actual data requests hit the API. Every tool result carries structured metadata, including source attribution, data freshness, and rate-limit cost, enabling agents to cite sources and plan request budgets. The server also ships with six workflow prompts that can become one-click flows in clients that support MCP prompts, such as market and macro research, sourced stock snapshots, sanctions and compliance screening, industry comparisons, climate/maritime/trade intelligence, and source discovery.
In terms of usage, the hosted MCP is recommended, with no installation required. In Claude, ChatGPT, and Codex, the Sugra API MCP server can be added from a directory: Claude web, desktop, mobile, Claude Code, and Cowork can be added via Anthropic's Connectors Directory and logged in with a Sugra account; ChatGPT and Codex can be added via OpenAI's Plugins Directory. Other MCP clients or manual setups can point to the hosted Streamable HTTP endpoint https://mcp.sugra.ai/mcp（永久别名 https://app.sugra.ai/mcp），使用 OAuth login or API key authentication with Authorization: Bearer sugra_xxx_.... Local installation runs via stdio (or self-hosted HTTP), authenticates with SUGRA_API_KEY, and a free API key can be obtained at app.sugra.ai/register, with the free tier at 50 requests per day.
The local package exposes eight gateway tools, and the hosted endpoint additionally provides three composite analysis tools on top of that (resolve_entity, get_snapshot, and get_timeseries); these tools rely on infrastructure credentials available only in the hosted deployment, so the stdio and self-hosted HTTP of pip install sugra-api-mcp always expose only the classic eight-tool gateway. For compliant entity queries (LEI / VAT, sanctions screening), sugra_entity_lookup and sugra_entity_screen can be used, and they are available across all transports.
Applicable scenarios include: market and macro research (such as comparing G7 inflation and central bank policy rates), stock snapshots and source citation, sanctions and compliance screening, industry comparisons, climate/maritime/trade intelligence, and data source discovery. Supported clients include Anthropic Claude (Claude Desktop, Claude Code, claude.ai), OpenAI GPT (ChatGPT via the MCP connector), Google Gemini (Gemini CLI, Gemini Code Assist), xAI (Remote MCP Tools in the xAI SDK and Responses API), IDEs (VS Code, Cursor, Zed, Cline, Continue.dev, Windsurf), and custom agents built on the Python or TypeScript MCP SDK.
In terms of limitations, the three composite tools exclusive to the hosted endpoint are not available in local stdio or self-hosted installations. ChatGPT sessions (where the client name in MCP initialize is openai-mcp, or where no client name is sent and the User-Agent starts with it) will list 145 typed tools instead of search_endpoints, describe_endpoint, call_endpoint, and fetch_data—these typed tools do not accept raw positional input (coordinates, addresses, postal codes). A fields projection accepts at most 32 paths, each path at most 256 characters and 16 dot segments; exceeding this returns projection_too_large; a single projection accesses at most 100,000 list items, runs for at most 5 seconds, and processes at most 2,000,000 characters of JSON response; the result itself has an 18,000-character limit. Shaping operations run on a separate pool of two worker threads, with at most 8 tasks running or waiting and at most 4 per single caller; if no free slot is found within 2 seconds, server_busy is returned.
