# Connect a financial research agent Use the public Streamable HTTP server: ```text https://api.seiche.info/openbb/mcp ``` Our connection needs no API key. Your agent platform may require its own account, model credentials and paid plan. Inspect its current tool list before enabling it. Start with this task: > Discover the available Financial Evidence datasets. Review USD funding and the > covered institution ESAF. Cite the source URL, observation date, unit and > availability for each observation. Keep the funding and institution sections > separate. Save the evidence and identify what needs human review. ## Claude / Cowork In Settings → Connectors, add a custom connector with the URL above. For Claude Code, run: ```sh claude mcp add --transport http financial-evidence https://api.seiche.info/openbb/mcp ``` This uses a custom connection. Official connector-directory publication is a separate process. [Claude instructions](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp). ## ChatGPT / Codex Use a custom Streamable HTTP MCP connection where your account and administrator permit it. For the Codex CLI: ```sh codex mcp add financial-evidence --url https://api.seiche.info/openbb/mcp ``` The repository also contains an installable Codex plugin. Universal directory publication remains pending; this guide does not claim a ChatGPT listing. [Plugin installation](https://beepboop2025.github.io/financial-evidence-skills/llms.txt) · [OpenAI submission documentation](https://developers.openai.com/plugins/deploy/submission). ## Microsoft Copilot Studio Add an MCP server tool to an agent using Copilot Studio's current MCP flow. Enter the URL above and use the unauthenticated connection option where available. Select the dataset discovery and research-query tools, test the task in the test pane, and inspect the cited result before publishing your agent. Tenant policy can restrict external servers. This setup has not been tested in your tenant. [Microsoft instructions](https://learn.microsoft.com/en-us/microsoft-copilot-studio/agent-extend-action-mcp). ## Flowise For an existing self-hosted Flowise 3.1.4 installation, use the [tested Start → Tool template](flowise/README.md). It retrieves one bounded financial research page without a model connection or financial-data API key. Four saved workflows passed through Flowise's authenticated APIs, including the exact download; browser UI import remains unobserved. Official Flowise is end-of-life and its upstream template intake is closed. This is a self-hosted compatibility download with [native proof](flowise/native-proof.json), not an official marketplace listing. [Maintainer announcement](https://github.com/FlowiseAI/Flowise/discussions/6727). ## Dify Use a remote MCP connection where supported, or evaluate the repository's native tool plugin in `integrations/dify`. Version 0.1.1 passed three bounded Dify Cloud remote-debug workflows. Its [Marketplace submission](https://github.com/langgenius/dify-plugins/pull/3278) remains under review; [package and validation scope](native-packages.md#dify) are available for evaluation. [Plugin source](https://github.com/beepboop2025/financial-evidence-skills/tree/main/integrations/dify). ## Haystack Use the existing Financial Evidence component from [`integrations/haystack`](https://github.com/beepboop2025/financial-evidence-skills/tree/main/integrations/haystack). Its upstream integration review is tracked in [PR 650](https://github.com/deepset-ai/haystack-integrations/pull/650). Source installation and upstream catalog acceptance are distinct states. ## TradingView companion Connect Financial Evidence alongside TradingView's official server only if you already have eligible TradingView access. TradingView uses OAuth and requires Essential or above; trials are excluded. Its URL is: ```text https://mcp.tradingview.com/mcp ``` For a narrowly scoped review, enable only `list_watchlists`, `get_watchlist` and `search_symbols` from TradingView. Do not enable `get_active_watchlist`: its documentation says it can activate or create a watchlist. Tool-name instructions are not an access-control mechanism; enforce the selection in your client's tool permissions where supported, otherwise leave TradingView disconnected. > Read my existing watchlist. Ask me to select an institution covered by > LiquiLens; do not infer a symbol-to-bank match. Retrieve that institution's > evidence and the USD funding context. Cite sources and dates, distinguish > market data from filing evidence, and produce a research note for review. This is a companion workflow in an MCP client, not a TradingView indicator or marketplace listing. [Official TradingView setup and tool behavior](https://www.tradingview.com/mcp/docs). ## Alpaca companion Keep Financial Evidence in a separate research step beside an existing Alpaca paper-trading project. Run the cited funding task above before reviewing a paper simulation. This kit does not connect a brokerage account, ingest credentials or submit an order. Alpaca's official MCP has broader capabilities; any account connection and tool permission must be chosen separately. [Alpaca MCP documentation](https://docs.alpaca.markets/us/docs/trading-api-mcp). ## Acceptance check 1. Retrieve one bounded research response and open its cited source. 2. Confirm unavailable values stay missing and observation dates remain visible. 3. Save the response with a useful research note. 4. Repeat on another working day and record whether it changed the review. The first three prove a useful setup. Repeat use by an independent user is the adoption signal. A synthetic probe, download or directory listing does not prove customer adoption.