{"protocolVersion":"0.2.6","name":"EvidInvest Research Agents","description":"Financial research agents combining a reasoning LLM (Claude on Amazon Bedrock) for intent understanding and narrative with deterministic valuation engines and cited SEC indexes for every figure — the model can never invent a number. Choose the writing model per request (default Claude Opus 5.5; any catalog key, e.g. sonnet-5.5, haiku-5.5, haiku-4.5 included, or none for engine output only — bring your own model); premium models carry a credit surcharge and fall back to a cheaper model, never to a charge you cannot cover. Runs that use a model need purchased credits (free starting credits cover data calls only; model 'none' is a data run). The valuation agent produces a full cited valuation report (DCF, fair-value range, margin of safety, growth history) for any listed US company. The SEC filing agent answers research questions with exact filing-section text, accession-numbered citations and supply-chain evidence — name the company in your request and the search is scoped to that filer; leave it unnamed and it runs across all filers, ranked by relevance only, and the answer says so. Pay-per-call credits, no subscription.","url":"https://mcp.evidinvest.com/a2a","version":"1.0.0","provider":{"organization":"Evid Invest AB","url":"https://evidinvest.com"},"documentationUrl":"https://evidinvest.com/developers","capabilities":{"streaming":false,"pushNotifications":false,"stateTransitionHistory":false},"securitySchemes":{"oauth2ClientCredentials":{"type":"oauth2","flows":{"clientCredentials":{"tokenUrl":"https://mcp.evidinvest.com/oauth/token","scopes":{"read":"Call research tools"}}},"description":"Two-legged OAuth. Get client credentials by POSTing your EvidInvest API key to /oauth/register."},"bearerApiKey":{"type":"http","scheme":"bearer","description":"EvidInvest API key (evid_sk_...) — create one at evidinvest.com/developers or via the MCP create_account tool."}},"security":[{"oauth2ClientCredentials":["read"]},{"bearerApiKey":[]}],"defaultInputModes":["text/plain"],"defaultOutputModes":["text/markdown","application/json"],"metadata":{"modelOptions":{"description":"Choose the model that writes the report per request via params.configuration.model (or message.metadata.model). Allowed: none, haiku-4.5 (fast, 1x), sonnet-4.5 (default, 3x), sonnet-4.6 (default, 3x), opus-4.5 (deep, 5x), opus-4.6 (deep, 5x), haiku-5.5 (fast, 0.11x), sonnet-5.5 (default, 2.2x), opus-5.5 (deep, 4.4x), fable-5.1 (deep, 11x), llama-3.3-70b (fast, 0.57x), deepseek-r1 (deep, 1.28x). Multiplier = cost of a typical final write relative to haiku-4.5 (included). Default: opus-5.5, falling back to sonnet-5.5 and then haiku-4.5 when credits or Bedrock availability don't cover it. Premium surcharge in credits per run (valuation/filing search): opus-5.5 +30/+17, sonnet-5.5 +11/+6, haiku-5.5 +0/+0, fable-5.1 +87/+48, opus-4.6 +35/+20, sonnet-4.6 +18/+10; haiku-4.5 and haiku-5.5 are included. If the surcharge exceeds your remaining credits, or the model is not enabled on Bedrock, the run falls back to sonnet-5.5, then haiku-4.5, and task.metadata.modelNote says why — you are only charged for the model that actually wrote. 'none' = no LLM, deterministic engine output only (bring your own model and pass metadata.skillId). Legacy values 'haiku' (= haiku-4.5) and 'sonnet' (= sonnet-4.5, +1 call unit) still work. AWS Marketplace subscribers pay model cost through the monthly llmusage line instead of credits. An unknown key is rejected with JSON-RPC error -32602 listing the allowed keys.","values":["none","haiku-4.5","sonnet-4.5","sonnet-4.6","opus-4.5","opus-4.6","haiku-5.5","sonnet-5.5","opus-5.5","fable-5.1","llama-3.3-70b","deepseek-r1"],"default":"opus-5.5"}},"skills":[{"id":"valuation-report","name":"Company Valuation Report","description":"Full valuation of a listed US company: snapshot, DCF with projections and WACC, bear/base/bull fair-value range, margin of safety, and multi-period growth rates. Give it a ticker (e.g. 'Value AAPL').","tags":["valuation","DCF","fair value","margin of safety","equity research"],"examples":["Value AAPL","Run a valuation report on MSFT","What is NVDA worth? Include margin of safety."],"inputModes":["text/plain"],"outputModes":["text/markdown","application/json"]},{"id":"sec-filing-search","name":"SEC Filing Research","description":"Answers research questions from SEC filings and earnings calls: exact section text with accession-numbered citations and sec.gov links, plus supply-chain relationships extracted from filings. Ask anything a 10-K can answer. Name the company you mean, and name the document when you know it — the form (10-K, 10-Q, 8-K), the company's own fiscal year, the section (Item 1A) and the quarter are read straight out of your request and used as filters, and a request that merely names a document (\"NVDA's latest 10-K\") is answered by returning that filing's sections in filing order rather than by ranking anything. A request that names its subject is scoped to that filer and is fast and precise; one that does not searches every filer, is ranked by relevance only, and the answer carries a quality caveat you should read before trusting it. Questions that are genuinely about many companies (\"which filers name NVIDIA as a supplier\") are welcome — just expect the caveat.","tags":["SEC","filings","10-K","earnings call","supply chain","citations"],"examples":["NVDA latest 10-K","What does NVDA's 10-K say about supplier concentration?","Risk factors in DELL's FY2027 10-K, Item 1A","What did NVDA management say on the Q2 FY2027 earnings call?","Search filings: MSFT commercial remaining performance obligation","Show AAPL's supply chain from filings","Which filers name NVDA as a supplier? (cross-company, caveated)"],"inputModes":["text/plain"],"outputModes":["text/markdown","application/json"]}]}