Revelata deepKPI MCP Server

官方

來自SEC申報的深度基本面數據，包含Bloomberg上找不到的營運KPI，專為您的金融AI代理打造。

文件

Revelata deepKPI MCP 伺服器 — 說明文件

基礎面數據，包含連 Bloomberg 都找不到的營運 KPI — 預先結構化、完整引用、代理程式立即可用。

透過 MCP 提供金融 AI 所需的深度基礎面數據

真正驅動分析的基礎面數據 — 可比銷售額、ARPU、門店數、積壓訂單、部門層級營收、管理層展望、非 GAAP 細項 — 都存在於 SEC 申報文件的文字、表格和圖表中。Revelata 的 deepKPI 將這些資訊提取並預先結構化為 KPI 時間序列，每個數據點都附有句子層級的引用來源，讓您和您的代理程式能夠信任這些數字。

我們將這些數據與底層的文字脈絡和評論相結合，讓您的 AI 代理程式成為您終於可以信賴的分析師。

您可獲得：

330 萬個公司特定 KPI 時間序列，涵蓋美國上市公司（S&P 500、400、600），以可查詢的時間序列形式結構化，提供超過 10 年的歷史數據並每日更新
自然語言 KPI 探索與語義搜尋 — 查詢「Olive Garden 客流量」或「AI 資本支出展望」，無需知道申報文件中的確切用詞，即可找到正確的指標
轉換為 Markdown 的 SEC 申報文件（10-K、10-Q、8-K），保留表格、圖表和結構 — 可完整檢索，讓代理程式能逐字引用管理層說法
主題式公司探索 — 透過從 10-K 衍生的摘要進行語義搜尋，根據公司實際業務而非僅依 SIC 代碼來發掘公司
隨需提供的公司摘要與部門細分，源自每家公司最新的 10-K

專為需要正確無誤的代理程式打造

我們為您的金融 AI 代理程式提供可信賴的脈絡。每個回傳的 KPI 數據點都帶有單位、規模、期間，以及指向確切來源句子的連結。每次申報文件檢索都回傳逐字的 Markdown，因此代理程式不會改寫重大事實。

代理程式技能也可取得

開源的代理程式技能，教導代理程式如何將這些工具用於常見工作流程 — 季節性分析、隱含指標推導、分析師報告壓力測試、基準比較等 — 可在 GitHub 找到。

免費試用

每位使用者每月 100 點免費額度。查詢和申報文件列表永遠免費；提取和搜尋呼叫則按結果計費。國際市場覆蓋（日本、中國、香港和新加坡）可供企業客戶使用。

設定說明

首先，建立一個免費的 Revelata 帳戶 — 註冊。

然後，按照以下說明，將伺服器與 Claude Desktop、Claude.ai 及類似客戶端搭配使用。

注意： OpenClaw 不使用我們的 MCP 伺服器；它使用 REST API 和 DEEPKPI_API_KEY。請參閱 skills/_common/deepkpi-api.md。託管 API 文件：deepkpi-api.revelata.com/docs。

基本設定（Claude）

複製以下 MCP 伺服器 URL https://deepkpi-mcp.revelata.com/mcp
Allow required capabilities — In Claude, enable Code execution and file creation by navigating to Settings → Capabilities on Claude.ai. For other clients, consult your documentation for connecting to MCP servers.
Add a custom connector —
Claude Desktop: Settings → Connectors → Add custom connector.
Claude.ai (web): Use Settings → Customize.
Configure the connector — Enter Revelata deepKPI under name and paste the server URL from step 3 exactly as copied. Click 'Add'.
Complete OAuth — Begin a chat or task. When prompted to authenticate, sign in with your Revelata account in the browser.
Begin using deepKPI data in your analyses!

MCP tools

The server exposes the following tools:

`query_company_id`

Looks up a company’s numeric identifier from a free-text name. Returns matches you can use as company_id in the other tools. For US companies this identifier is the SEC CIK.

Parameters: company_name (string), num_of_res (optional, default 5) — max number of matches to return.

Credits: None — free.

`list_kpis`

Lists all KPIs available for a given company, organized by category. Use this as the first step before searching: the returned KPI names help you write precise queries for search_kpis, which saves credits.

Parameters: company_id (string), source (optional) — omit to include all filing types, or set to 10-K, 10-Q, or 8-K to filter.

Credits: None — free.

`search_kpis`

Semantic search over KPIs for one company — finds items most relevant to your natural-language query. Each result costs one credit (cap 15 results per call; default 3 — increase only when needed). Best practice: run list_kpis first, then search using exact KPI names from that list when possible instead of requesting many broad results.

Parameters: company_id (string), query (string), num_of_res (optional, default 3), source (optional) — same filing-type filter as list_kpis; omit to search across all filing types.

Credits: Charged per result — check your balance or add credits at AI credits.

`company_summary_search`

Semantic search over all companies’ 10-K-derived summaries — returns the best-matching tickers/CIKs for a natural-language query (thematic discovery: “companies that …”, “who operates in …”). 1 credit per company returned (cap 15 via top_k_companies); empty results cost nothing.

Parameters: query (string), top_k_companies (optional, default 10, max 15).

Credits: 1 per company in the response — check your balance at AI credits.

Agent workflow: the company-summary skill — use for thematic lists; for a named company’s profile use get_company_summary / get_company_segments first (Workflow A in that skill).

`get_company_summary`

Returns a narrative company summary (derived from the company’s latest 10-K). Primary source for “what does this company do?” — not for pulling numeric KPI time series (use search_kpis for that). May follow company_summary_search when a thematic hit needs full prose.

Parameters: company_id (string) — for US companies, the SEC CIK (same as for list_kpis / search_kpis). If you only have a name, call query_company_id first.

Credits: 3 per successful call.

Agent workflow: the company-summary skill.

`get_company_segments`

Returns a structured segment breakdown (derived from the company’s latest 10-K). Use for high-level segment mix and narrative structure; for detailed segment metrics over time, prefer list_kpis + search_kpis.

Parameters: company_id (string) — SEC CIK for US companies; resolve with query_company_id if needed.

Credits: 3 per successful call.

Agent workflow: the company-summary skill.

`list_sec_filing_markdowns`

Lists SEC filings available as markdown for a given CIK. Use this to discover valid acc_no and seq_no values before fetching content. Agent workflow: the filing skill (tool-first, before web/SEC.gov).

Parameters: cik (int), form_type (optional string, e.g. 10-K), start_date (optional YYYY-MM-DD), end_date (optional YYYY-MM-DD).

Credits: None — free.

`get_sec_filing_markdown`

Fetches the markdown content for a specific SEC filing.

Parameters: cik (int), acc_no (string), seq_no (optional int, default 1).

Credits: 10 per successful call.

Critical quoting rule: When the user asks “what did they say”, “what comments were made”, “exact language”, or similar, you MUST return verbatim quotes/snippets from the markdown (with clear snippet boundaries). Do not paraphrase by default. Full rules: the filing skill.

Troubleshooting

Authentication issues

Symptom What to try

Redirect loop or “access denied” Confirm you’re signed into the expected Revelata account.

Connector works but search / summary / filing markdown calls fail You may be out of credits. list_kpis, query_company_id, and list_sec_filing_markdowns are free. search_kpis uses 1 credit per result; company_summary_search uses 1 credit per company returned; get_company_summary and get_company_segments each use 3 credits on success; get_sec_filing_markdown uses 10 credits on success. Check your balance at AI credits. Note that all accounts receive 100 free credits each month.

Symptom	What to try
Redirect loop or “access denied”	Confirm you’re signed into the expected Revelata account.
Connector works but search / summary / filing markdown calls fail	You may be out of credits. `list_kpis`, `query_company_id`, and `list_sec_filing_markdowns` are free. `search_kpis` uses 1 credit per result; `company_summary_search` uses 1 credit per company returned; `get_company_summary` and `get_company_segments` each use 3 credits on success; `get_sec_filing_markdown` uses 10 credits on success. Check your balance at AI credits. Note that all accounts receive 100 free credits each month.

Network and TLS

Use HTTPS only for the MCP endpoint.
Allow outbound HTTPS to your MCP host; some corporate networks interfere with long-lived streaming — try another network to test.
If auth hangs, try toggling VPN.

Tools don’t appear

Restart Claude after changing the connector.
Re-add the connector and complete OAuth again.
Confirm your client supports remote MCP with Streamable HTTP (not only legacy SSE docs).

OpenClaw vs MCP

Runtime	How you use deepKPI
Claude + MCP	Connector URL `…/mcp`, OAuth — tools `query_company_id`, `list_kpis`, `search_kpis`, `company_summary_search`, `get_company_summary`, `get_company_segments`, `list_sec_filing_markdowns`, `get_sec_filing_markdown`.
OpenClaw	Install `revelata-deepkpi`, set `DEEPKPI_API_KEY`, call REST per `skills/_common/deepkpi-api.md`.

Do not paste the MCP URL into OpenClaw as a substitute for the REST skill flow.

Get the Agentic Skills bundle, too

Agent instructions and OpenClaw install:

curl -fsSL https://raw.githubusercontent.com/revelata/deepkpi-agents/main/install.sh | bash

詳細資訊：README.md。

仍然卡住？

請立即聯繫 [email protected]！