SOTA2 MCP connector =================== Find public AI company and product information, solution leaderboards by industry and task, company revenue rankings, and research benchmarks, datasets and papers. Connection ---------- Server URL: https://www.sota2.com/mcp Transport: Streamable HTTP Authentication: No sign-in or API key is required for these public, read-only tools. In Claude, open Customize > Connectors > Add custom connector. Enter SOTA2 and the server URL, select No sign-in, add the connector and connect it. Other remote MCP clients can use the same URL and transport. Refresh the connection after a tool schema update. In ChatGPT, open Plugins > Add > Add custom MCP server. Enter the server URL, select No authentication, create and install the personal plugin, then select it in a new Work chat. Custom connections depend on account/workspace settings. Public directory listings are coming soon. Setup: https://www.sota2.com/with-ai Available tools --------------- search_companies: Resolve a named company to its ID or slug. get_company: Read profile, products, compliance claims, customers, deployment, funding, solution ranking appearances and revenue dimensions. list_industries / list_tasks: Discover solution ranking scopes and descriptions. get_solution_rankings: Read an industry task directory or an exact task leaderboard. list_company_leaderboards / get_company_leaderboard: Read revenue/country lists. list_research_domains / search_research: Discover research fields and entities. get_research_entity: Connect domains, tasks, datasets, benchmarks and papers. get_research_benchmark: Read metric-specific reported result rows and source papers. Example conversations --------------------- "Find Abridge and its products." "Does the second product have a reported SOC 2 claim? What integrations are listed?" "Show the healthcare Draft Documents leaderboard, top two results." "Next page. Compare that page's second company with the first company's products." "Show China's AI company revenue leaderboard, then the next page." "Find OmniDocBench v1.5, Text Edit Distance, top two reported result rows." "Only PaddleOCR-VL-related methods, two rows per page. Next page." "Show this result's reporting paper and available code." Reading results and follow-ups ----------------------------- Use returned IDs/slugs and retain the industry/task or benchmark version/split. Revenue-only companies may have no catalog slug or profile URL. Reuse their returned id, or a leaderboard row's companyId, to retrieve company data; missing catalog products/compliance stay unknown. Do not invent a company profile page. An ambiguous verification request needs clarification: identity/KYC, document authenticity and factual claims are different tasks. Solution positions apply to one industry/task. An industry directory is not a universal company ranking. Company counts overlap across tasks; adding those counts does not give a distinct-company total. Check product evidence and caveats against narrower needs such as a document type or integration. Research rankings preserve each result ID, method name, configuration and global position. Ties remain ties. A top-N request returns N individual rows; different reports or method variants are not merged into a family. For method-specific results, supply the method filter before pagination and follow results.nextCall. The filter is a case-insensitive substring match over the complete ranked benchmark. Changing the scope, metric or filter starts a new first page. Keep metric units and higher/lower-is-better direction, benchmark versions/splits, reporting and origin papers separate. Read quality warnings. Invalid percentage values are flagged and excluded, not silently rescaled. These are reported research results; a range check is not independent scientific verification. Company compliance fields are reported claims with statuses, not independent audits. Missing information is unknown. Revenue, ARR, run-rate, estimates and forecasts retain their basis, period, confidence and source. Cite returned URLs and dates, and disclose preview/snapshot sources and unavailable upstream data. The server is stateless: clients send the selected IDs and filters on every call. Collections use nextOffset; research discovery/entity lists use API page numbers. Each company collection paginates independently. No write tools, private account data, credentials, payments or booking actions are available through this connector. Lookup outcomes and troubleshooting ---------------------------------- No matches, missing industry/task combinations and ambiguous selections are normal results, not tool failures. They include a status such as not_found, no_matches, needs_scope or needs_selection, plus available discovery steps. They do not imply that no company serves a need. Another scope is never silently substituted. Benchmark display names return canonical discovery choices without selecting a version/split. Actual upstream API failures and invalid parameter formats remain tool errors. Dataset details use the canonical dataset-family slug in benchmark.dataset.familyLookup (for example, omnidocbench), not the dataset's version slug (omnidocbench-v1-5). Family details include multiple versions; the selected benchmark version/split stays separate. If a service fails, report the unavailable source rather than presenting an empty result as a successful lookup. Support ------- Website: https://www.sota2.com Questions and corrections: leo@pageonelab.com Privacy: https://www.sota2.com/mcp-privacy Terms: https://www.sota2.com/mcp-terms