Affinity
Search your CRM, prep for meetings and update deal records from your AI assistant.
Pull public-company financials, filings, ratios and ownership.
Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.
Paste it into any MCP client. Setup docs
The Fiscal.ai MCP server lets Claude, ChatGPT or any MCP-compatible agent query Fiscal.ai, a public-company financial data provider. An agent can pull a standardized income statement, chart a valuation multiple, or open a filing at the page a number came from. Sign-in is OAuth or an API key, and everything the connector reaches is read-only market data.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
api_docs or execute_code. Everything we know about the sandbox comes from a tool-schema manifest published for Microsoft Copilot. A reader who follows the documentation link alone learns that a code executor exists only by inference.isNonConsequential with confirmation type None. Defensible for read-only tools; worth knowing before an agent runs generated JavaScript unattended./mcp/sse; Fiscal.ai documents /mcp over streamable HTTP as current. Both answer, but register the newer one — Fiscal.ai notes Copilot Studio no longer supports SSE. This page's frontmatter mirrors the directory record, so it carries the /sse URL; the prose and the connect instructions use the current one.readOnlyHint or destructiveHint annotations were readable. Every tool description here is Fiscal.ai's own, taken from the manifests it publishes unauthenticated.fscl identifier and FIGI carry no such gate.In a sandbox on Fiscal.ai's own servers, never on your machine. The tool schema Fiscal.ai publishes describes a network-isolated sandbox with a 30-second budget, and says code must arrive as a single async arrow function. Nothing is installed locally and no code touches infrastructure you administer. Fiscal.ai does not name the sandbox technology.
Fiscal.ai states it cannot. Its published tool schema calls the sandbox network-isolated, so generated JavaScript reaches Fiscal.ai's own helper functions rather than the open internet. That is a containment claim we could not test without authenticating and executing code, which this catalogue does not do. Treat it as documented design, not verified behaviour.
Fiscal.ai's API reference documents 48 GET endpoints across roughly seventeen groups, counted on 2026-08-19. They cover company profiles, as-reported and standardized statements, ratios, segments and KPIs, ownership, filings, transcripts, news and fund letters. That is the whole reason the connector ships two tools instead of 48 — the definitions would crowd out the context window.
Yes, in substance. Every one of the 48 documented endpoints is a GET, so generated code can query market data but cannot write to anything at Fiscal.ai. The execute_code name is a strong write verb in general, and here it is honest only because the sandbox is network-isolated and the API behind it is read-only.
It is gated by your Fiscal.ai plan rather than metered per call. Fiscal.ai states MCP enforces the same entitlements as its REST API, so a free key gets 250 requests a day, a 50-per-minute burst and only 100 companies. No credit is spent by the connector itself, and seeing a tool does not mean your plan can reach it.
Eleven, and they map to data domains rather than to read and write. Fiscal.ai's authorization server advertises financials, financials_sourcing, segments_and_kpis, stock_quotes, filings, adjusted_numbers, news, ir_events, ownership, fund_letters and fiscal_earnings_calendar. We read them live on 2026-08-19. There is no scope specific to code execution, because every scope is read-shaped already.
Use the streamable HTTP endpoint at api.fiscal.ai/mcp. Fiscal.ai documents that URL as the current one and describes the /mcp/sse address as a legacy endpoint kept for older clients. Anthropic's directory snapshot still carries the SSE form. Fiscal.ai specifically notes Microsoft Copilot Studio dropped SSE support, so register the newer URL there.
.md suffix form returns a Next.js error shell, so this was read as HTML · retrieved 2026-08-19isNonConsequential annotation, GetPrivateData and DataTransform handling tags, and the terms and privacy URLs) · retrieved 2026-08-19allCompanies parameter) · retrieved 2026-08-19www-authenticate challenge. The root form returns a byte-identical 357-byte body; the path-insert form returns 401) · retrieved 2026-08-19issuer, the eleven scopes_supported entries, PKCE S256, grant types and registration endpoint) · retrieved 2026-08-19https://api.fiscal.ai/mcp/sse returns HTTP 401 with www-authenticate: Bearer realm="OAuth", resource_metadata="https://api.fiscal.ai/.well-known/oauth-protected-resource/mcp/sse", error="invalid_token" · retrieved 2026-08-19docs.fiscal.ai/robots.txt (fetched 2026-08-19) allows all user agents and carries no Content-Signal axes. fiscal.ai/robots.txt could not be read: the apex host returns a Vercel bot challenge (HTTP 429) to curl with and without a browser user agent, which is a bot-mitigation response rather than a policy signal · retrieved 2026-08-19llms.txt is published) · retrieved 2026-08-19/mcp/sse endpoint) · retrieved 2026-08-16Connect Fiscal.ai once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.