EMMA MCP
EMMA MCP — 为 Blaylock Van 定制的市政金融代理
纽约大学斯特恩商学院咨询顶点项目 · 2026 年春季 · Marco Figueroa, Justin Ganjian, Andrew Kay, Isaac Mizrahi, Roshan Raja · Marciano 教授
这是什么
这是一个 自定义 MCP (Model Context Protocol) 服务器,让 Claude 可以直接从 MSRB 的 EMMA 网站读取市政债券新发行市场数据。这是我们顶点项目分级阶梯中的第 4 层示例——一个由公司拥有的、针对任何供应商都未覆盖的数据源的连接器。
EMMA(由市政证券规则制定委员会运营的电子市政市场准入网站)是美国每笔市政债券销售的官方公共记录簿。新发行日历、官方声明 (Official Statements)、CUSIP 级详细信息、近期交易价格:所有这些都在 EMMA 上。它没有公共 API。
此 MCP 公开了 18 个工具 和 2 个资源(一个 CLAUDE.md 规则文件和一个 EMMA_ABBREVIATIONS.md 术语表),Claude 在连接时会自动加载它们。安装后,银行家可以用自然语言向 Claude 提问,例如“拉取所有未偿还的 Rady Children's Hospital 债券并下载最新的 OS”,Claude 会在约 20 秒内执行完整的链条——搜索 → 发行人查询 → 发行详情 → PDF 下载 → 文本提取。
有关完整介绍以及它对 Blaylock 的重要性,请参阅 USAGE.md → 为什么存在此工具。
Related MCP server: IsoFinancial-MCP
先决条件
你的笔记本电脑需要具备四样东西。如果缺少任何一项,下面的安装说明会涵盖它们。
要求 | 说明 | 安装检查 |
Claude Desktop | Anthropic 的原生 macOS/Windows 客户端(MCP 主机) | |
Python 3.10+ | 服务器运行环境 | Mac: |
uv | Python 包管理器(自动处理虚拟环境) |
|
Playwright Chromium | 服务器用于驱动 EMMA 的无头浏览器 | 通过下方的一条命令安装 |
安装 — Mac
# 1. Unzip this folder somewhere sensible
cd ~/Downloads
unzip emma-mcp.zip -d ~/emma-mcp
cd ~/emma-mcp
# 2. If you don't have uv:
brew install uv
# 3. Create the venv + install dependencies (one command)
uv sync
# 4. Install the headless browser Playwright drives
uv run playwright install chromium
# 5. Smoke-test the server (optional — confirms it boots)
uv run python server.py
# You should see the server start and wait on stdin.
# Press Ctrl+C to stop.将其连接到 Claude Desktop (Mac)
打开 Claude Desktop → Settings → Developer → Edit Config。这将打开
claude_desktop_config.json。在
mcpServers对象内添加一个emma条目。使用uv sync刚刚创建的.venv/bin/python的 绝对路径 以及server.py的 绝对路径:
{
"mcpServers": {
"emma": {
"command": "/Users/YOURNAME/emma-mcp/.venv/bin/python",
"args": ["/Users/YOURNAME/emma-mcp/server.py"]
}
}
}保存,完全退出 Claude Desktop(使用
Cmd+Q,而不仅仅是关闭窗口),然后重新打开。在新的聊天中,编辑器中的锤子图标现在应该列出 EMMA 工具。这就是成功的标志。
安装 — Windows
# 1. Unzip this folder, e.g. to C:\emma-mcp
cd C:\emma-mcp
# 2. If you don't have Python: install from python.org (3.10 or later).
# Check "Add Python to PATH" during install.
# 3. Install uv
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
# 4. Create the venv + install deps
uv sync
# 5. Install Playwright's Chromium
uv run playwright install chromium
# 6. Smoke-test
uv run python server.py
# Ctrl+C to stop.将其连接到 Claude Desktop (Windows)
在文本编辑器中打开
%APPDATA%\Claude\claude_desktop_config.json。(将该路径粘贴到资源管理器的地址栏中——Windows 会将%APPDATA%解析为C:\Users\YOU\AppData\Roaming。)在 JSON 路径中使用 正斜杠或双反斜杠 —— Windows 的单反斜杠会破坏 JSON:
{
"mcpServers": {
"emma": {
"command": "C:/emma-mcp/.venv/Scripts/python.exe",
"args": ["C:/emma-mcp/server.py"]
}
}
}保存,完全退出 Claude Desktop(右键点击托盘图标 → 退出),重新打开。
检查编辑器中的锤子图标,查看是否有 EMMA 工具。
首次提示 — 60 秒验证
打开一个新的 Claude Desktop 聊天并粘贴:
使用 EMMA MCP 拉取下周即将发布的新发行日历。仅筛选纽约州。返回一个包含发行人、面值、日期和行业的整洁表格。
你应该看到 Claude 调用 get_new_issue_calendar 并传入 state="NY",在几秒钟内返回约 10-25 行数据,并将其格式化为带有 EMMA URL 的表格。如果这有效,其他所有工具的工作方式也是一样的。
想要更令人印象深刻的演示,请尝试:
在 EMMA 上找到 Rady Children's Hospital,拉取他们的未偿还债券,获取最近一次发行的详细信息,下载官方声明,并总结前 10 页的评级语言和收益用途。
这会在约 20 秒内端到端地运行完整的五工具链。
更多演示提示请参阅 USAGE.md。
故障排除
重启后 Claude Desktop 不显示 EMMA 工具。
确认 JSON 有效(
python3 -m json.tool < claude_desktop_config.json— 无输出即为良好)。确认配置中的两个路径都是绝对路径且确实存在。相对路径和
~不起作用。完全退出 Claude Desktop(
Cmd+Q/ 托盘 退出)——关闭窗口会使其保持运行。检查 Claude Desktop 的 MCP 日志,路径为
~/Library/Logs/Claude/mcp*.log(Mac) 或%APPDATA%\Claude\logs\mcp*.log(Windows),查看具体错误。
playwright install chromium 挂起或失败。
首次安装需下载约 150 MB,在家庭网络上需要 1-2 分钟。请耐心等待。
如果在公司代理后方,请在运行前设置
HTTPS_PROXY和HTTP_PROXY。
Claude 调用工具时返回“EMMA 免责声明循环”或超时。
EMMA 在首次访问时会设置 CUSIP 免责声明 cookie。服务器会预先植入它 (
Disclaimer6=msrborg),但极少数情况下 EMMA 会更改 cookie 名称。重启 Claude Desktop 以重置浏览器上下文。如果问题仍然存在,可能是 EMMA 宕机了 — 请在普通浏览器中检查https://emma.msrb.org。
返回的表格显示 cusip_hash: ABC123… 而不是 9 位数的 CUSIP。
这是故意的,不是错误。EMMA 将 CUSIP 渲染为图像标签(CGS 许可规则),因此服务器无法从 HTML 中读取它们。官方声明 PDF 是权威的 CUSIP 来源。 使用
download_official_statement然后使用extract_pdf_text— 封面和内封面包含真实的 CUSIP。
“Rady Children's”搜索找到了俄亥俄州富兰克林和宾夕法尼亚州匹兹堡的儿童医院,但没有找到 Rady。
这是 EMMA 已知的怪癖,不是服务器错误。Rady 的债券归档在加州管道(CHFFA / CPFA)下,仅在括号内的义务人部分包含 "RADY CHILDRENS",EMMA 的快速搜索确实索引了它,但索引较弱。尝试
emma_quick_search {query: "rady"}或直接搜索管道:search_issuers_by_state {state: "CA", contains: "health care facilities"}。完整的义务人查询手册请参阅USAGE.md→ 管道速查表。
服务器启动但 Claude 说“没有可用工具”。
你可能编辑了
server.py并且存在 Python 语法错误。在终端中运行uv run python server.py— 任何错误都会打印到 stderr。修复后重启 Claude Desktop。
此文件夹中的内容
文件 / 文件夹 | 说明 |
| MCP 服务器本身 — 约 1,500 行,Python,Playwright。18 个工具,2 个资源。 |
| Blaylock 合规规则。作为 MCP 资源 |
| EMMA 的电报式发行人名称缩写(HOSP, AUTH, SR, CMNTY 等)+ 管道速查表。作为资源 |
| 本文件。 |
| 详尽参考:18 个工具、规范的工作流链、银行家提示示例、注意事项。 |
| Python 依赖清单。 |
| 幻灯片准备期间使用的单次驱动程序 — 直接调用 MCP 以便截图输出。运行服务器不需要。 |
| 辅助脚本,为顶点项目的第 2 层 (Cowork) 演示批量下载 OS PDF。运行服务器不需要。 |
| 测试期间通过 MCP 下载的示例官方声明。保留或删除均可 — 该文件夹会在首次调用 |
一句话版本
我们构建了没有任何供应商提供的 EMMA MCP,通过与每个市场连接器相同的协议将其连接到 Claude,并用银行家可以阅读和编辑的纯 Markdown 规则文件对其进行管理。这就是当一家公司真正拥有自己的工具时,第 4 层所呈现的样子。
Available Tools
18 toolsclassify_sectorA
Classify a muni bond issue description into a sector using the same keyword rules the calendar tool uses. Useful to show Marciano the classifier is deterministic, not a black box.
Inputs: description (required).
download_official_statementA
Download an Official Statement PDF from EMMA to local disk. Pass either a direct PDF URL or an IssueView URL (the tool will resolve it to the PDF link). Returns file path + size. Use when the banker wants the OS to attach to a deck or read offline.
Inputs: url (required), filename.
emma_quick_searchA
START HERE when the user names a specific obligor, issuer, or keyword. Uses EMMA's global quick search (the same box in EMMA's header) which indexes BOTH long-form and obligor-in-parens names — so 'Childrens Hospital' finds 'CALIFORNIA HEALTH FACILITIES FINANCING AUTHORITY (LUCILE SALTER PACKARD CHILDRENS HOSPITAL AT STANFORD)' without needing the conduit-issuer chase. Returns matching issuers (across all states) and matching specific issues. Each issuer row has a URL you can feed into get_issuer_outstanding_bonds.
Inputs: query (required), limit.
extract_pdf_textA
Extract text from a previously-downloaded OS PDF so you can pull financials, debt service tables, obligor financials, rating language, bond purpose, sources-and-uses, etc. into an Excel model. Works best on OS PDFs saved via download_official_statement.
Inputs: path (required), pages, max_chars.
get_calendar_summaryA
Aggregate the new-issue calendar by a grouping dimension. Returns count and total par per group, sorted by par descending. Perfect for slides: 'NY leads this week at $1.2B across 8 deals; hospitals and schools dominate the sector mix.'
Inputs: group_by, top_n.
get_issue_detailsA
Full IssueView page for a given IssueView URL — per-CUSIP detail (par at issuance, coupon, maturity, initial offering price, current LT rating, price, yield, and ratings from Fitch, KBRA, Moody's, S&P), the Official Statement PDF link, and recent trade activity. This is the single most information-dense page on EMMA for modeling a specific deal.
Inputs: issue_url (required).
get_issuer_outstanding_bondsB
List an issuer's outstanding bond issues from their EMMA profile — issue description, dated date, maturity range. The most reliable drill-down on any issuer.
Inputs: issuer_id, issuer_url, limit.
get_issuer_profileA
Full issuer profile: name, outstanding issues, attempted fetch of Official Statements, Pre-Sale docs, Continuing Disclosures, and Recent Trades. EMMA loads the latter via tab interaction; empty sections are reported honestly rather than guessed at.
Inputs: issuer_id, issuer_url.
get_market_pulseA
Snapshot of this week's muni primary market: total par, deal count, top 5 states, top 5 sectors, top 5 lead managers if available, competitive-vs-negotiated split. Built entirely from the live new-issue calendar — one call, boardroom-ready.
Inputs: no named parameters.
get_new_issue_calendarA
Upcoming municipal bond issues from MSRB EMMA — the primary competitive-intel tool for any muni banker. Each row is enriched with an inferred sector classification. Multi-axis filters: state, sector, tax_status, bank_qualified, sale_method, description_contains, min_principal, max_principal. Use this to answer questions like 'what hospital deals over $100M are pricing this week?'
Inputs: state, sector, tax_status, bank_qualified, +6 more.
get_official_statement_pdfA
Given an EMMA IssueView/Details/P##### URL, return the direct Official Statement PDF URL (and issue title). This is the canonical primary-source OS — use it before quoting coupons, maturities, or call features.
Inputs: issue_url (required).
get_recent_annual_reportsA
Recent issuer annual financial reports filed as continuing disclosures — the source for post-issuance updated obligor financials (revenue, operating income, debt-service coverage).
Inputs: limit, issuer_contains.
get_recent_continuing_disclosuresA
Recent continuing-disclosure filings (audited financials, material events, rating changes, quarterly operating data). Each row has a direct link to the filed PDF. This is where post-issuance credit work lives — ratings today, not at issuance; updated obligor financials for modeling.
Inputs: limit, disclosure_contains, issuer_contains.
get_recent_official_statementsA
List the EMMA Recent Official Statements grid — the 10 most recently filed OS documents across the muni market, with issuer, series, dated date, and a link back to the issuer. Feed a row's issue_url into get_official_statement_pdf to grab the PDF URL.
Inputs: limit.
get_recent_preliminary_statementsA
Recent Preliminary Official Statements (POS) — pre-pricing documents filed right before a deal comes to market. This is the leading edge of the new-issue pipeline. Each row links to the filing PDF and the issuer page.
Inputs: limit, issuer_contains.
get_recent_salesA
Most actively traded muni CUSIPs right now from EMMA's /TradeData grid — description, coupon, maturity, high/low price & yield, trade count, total trade amount, and a link to each security's EMMA detail page. Use this to gauge live-market color for comparable deals.
Inputs: limit.
get_security_detailsA
Pull security-level fields for a 9-digit CUSIP: issuer, coupon, maturity, dated date, par. CUSIPs that are redeemed / matured / pre-refunded return a clear 'no records' response — try a currently-outstanding CUSIP from get_issuer_outstanding_bonds.
Inputs: cusip (required).
search_issuers_by_stateA
List municipal issuers for a state (all 9k+ rows, not just page 1) and filter by name. Feed the issuer_id into get_issuer_outstanding_bonds or get_issuer_profile to drill down. IMPORTANT: the contains filter auto-expands your query into every EMMA abbreviation variant (hospital→HOSP, authority→AUTH, children→CHLDN, senior→SR, community→CMNTY, etc.) — always search in natural English, never in the telegraphic form. For hospital/senior-living/charter-school obligors, search the CONDUIT, not the obligor: e.g. CHLA bonds are filed under CALIFORNIA PUB FIN AUTH HEALTH CARE FACS REV — found by searching contains="health care facilities". See the emma://rules/EMMA_ABBREVIATIONS.md resource for the full variant map and conduit cheat sheet.
Inputs: state (required), contains, limit.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 18 tools
Each tool has a distinct purpose with detailed descriptions. For example, 'download_official_statement' vs 'get_official_statement_pdf' vs 'extract_pdf_text' are clearly separated. No two tools overlap in functionality; agents can unambiguously choose the right tool.
Names follow a consistent verb_noun pattern (e.g., 'classify_sector', 'download_official_statement', 'get_calendar_summary'). While not all use the same verb (mix of 'get_', 'search_', 'extract_'), the structure is uniform and clear, earning a 4 for minor deviation from a single prefix.
With 18 tools, the server covers a wide range of municipal bond data operations without being excessive. Each tool addresses a specific need, from searching issuers to extracting PDF text, making the count well-scoped for the domain.
The tool surface is comprehensive, covering search, issuer details, calendar, documents, and market data. Minor gaps exist, such as lack of historical trade analytics or advanced filtering, but core workflows are fully supported.
Maintenance
Related MCP Connectors
MCP server giving Claude AI access to 22+ NYC public-record databases for real estate due diligence
The Octagon MCP server provides specialized AI-powered financial research and analysis by integrating with the Octagon Market Intelligence API. It enables users to analyze public market data (SEC filings, earnings transcripts, financial metrics, and stock data for 8000+ companies), private market data (3M+ companies, 500k+ funding rounds, 2M+ M&A/IPO transactions), and conduct deep research including web scraping capabilities. The server also features autonomous research agents that search hundreds of sources and return fully cited reports in approximately one minute.
Hosted MCP server for live public-data APIs and Skills for AI agents.
75 MCP tools: SEC financials, FRED economics, IRS 990, FDA, FX, UK Companies House.
Related MCP Servers
- AlicenseAqualityAmaintenanceMCP server for the BoondManager API, enabling Claude to search, create, and modify records across 36 domains with 158 tools.238423 npm21Apache 2.0
- AlicenseNot gradedqualityDmaintenanceAn enhanced MCP server providing comprehensive financial market data endpoints for quantitative trading analysis, including SEC filings, FINRA short volume, earnings, news, and Google Trends. It features meta-tools that consolidate multiple data sources into single efficient calls, optimizing for AI agents with iteration budgets.8MIT

city-data-mcpofficial
AlicenseAqualityDmaintenanceAn MCP server that gives Claude deep access to US public data -- demographics, economics, crime, employment, weather, housing, transit, schools, budgets, and more across 30+ cities for government intelligence workflows.28MIT- FlicenseNot gradedqualityDmaintenanceA production-grade MCP server enabling Claude to perform comprehensive NJ real estate workflows including property search, valuation, neighborhood intelligence, investment analysis, and agent tools via 20 tools and 15+ data sources.-