Seiche, LiquiLens and Undertow source data

io.github.beepboop2025由 GitHub 账号 beepboop2025 发布v1.1.0更新于 Oct 24, 2026

Funding histories, bank filings and settlement data with source receipts, units and capture clocks.

发布方自己的服务：Source

已验证Streamable HTTP可网页运行金融数据与分析

服务详情

标识符: io.github.beepboop2025/financial-source-data
传输: Streamable HTTP
运行时: Web executable
信任: Verified
发布于: Sep 26, 2026

安全提示

MCP 服务会接收对话中的工具调用请求，并可能执行外部操作。启用前请先了解该服务能做什么，并只授予它所需的权限。

注册中心已验证此条目所属的 GitHub 账号 beepboop2025。这并不表明该账号就是产品方。

SourceWeft 的检查只涵盖条目是否完整、端点是否响应；不审查服务器的代码，也不审查其工具的行为。

在 README 中添加徽章

向读者展示该 MCP 服务已收录于 SourceWeft，徽章链接到本页。

[![Listed on SourceWeft](https://sourceweft.com/badge/mcp/io.github.beepboop2025%2Ffinancial-source-data.svg)](https://sourceweft.com/mcp/io.github.beepboop2025%2Ffinancial-source-data?utm_source=readme&utm_medium=badge)

概览

AI 生成的概览

托管型只读研究服务器，提供 LiquiLens、Seiche 与 Undertow 的源数据：融资历史、银行申报与结算观测，并附来源凭证。

可以做什么: 发现数据集、指标与已跟踪机构，并查询监管机构标识
读取带引用的观测值、已捕获的历史版本与描述性序列摘要
查看融资、机构集中度与链上结算分析及其输入引用
读取研究工作区的覆盖范围、来源新鲜度与缺口
适用场景: 适合需要带引用、带日期的银行资产质量、融资环境或结算数据证据的场景，并要求明确保留覆盖缺口与局限。适用于研究与复核工作，不用于交易或授信决策。
运行要求: 远程 Streamable HTTP 端点；公开只读工具无需本地安装、账号或 API 密钥。需要能访问该托管端点的网络。

安装前请注意

公开研究为只读，不执行交易，但可能适用公平使用限制，且模型服务商可能另行收费。覆盖范围有限且并非全量普查；过期或未经验证的输入会被扣留，结果为筛查而非评级或投资建议。

安装

在 SourceWeft 中

打开控制台中的 Seiche, LiquiLens and Undertow source data，将其添加到工作区。
为需要使用其工具的对话启用该服务。

Web executable，通过 Streamable HTTP。远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体，或 SourceWeft 对话中。智能体会读取该服务器的安装指南，先向你展示发布者、端点和登录方式，经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.github.beepboop2025%2Ffinancial-source-data/install.md，并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前，先告诉我它的发布者、端点和登录方式，等我确认。不要把 API 密钥写进命令或我会提交的文件；问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "seiche-liquilens-and-undertow-source-dat": {
      "type": "http",
      "url": "https://api.seiche.info/api/v2/research-data/mcp"
    }
  }
}

README

LiquiLens MCP — the Failure Radar as agent tools

Connect and ask one question

Hosted Streamable HTTP MCP; no local server or API key for public read-only tools. Add the products you need to Claude Code:

claude mcp add --transport http liquilens https://api.liquilens.in/mcpclaude mcp add --transport http seiche https://api.seiche.info/mcpclaude mcp add --transport http undertow https://api.seiche.info/undertow/mcp

claude mcp add --transport http liquilens https://api.liquilens.in/mcpclaude mcp add --transport http seiche https://api.seiche.info/mcpclaude mcp add --transport http undertow https://api.seiche.info/undertow/mcp

Try this first: What changed in ESAF Small Finance Bank's NPAs? Cite the filing, reporting period and missing disclosures.

Try it in your browser · Cursor configuration · Codex commands · Free three-product starter kit and skill

For a connected review, ask for institution evidence → funding context → BTC exit estimates, with each source date, coverage gap and limitation kept separate. Client/model costs and fair-use limits may apply. Research calls do not execute trades.

Endpoint: https://api.liquilens.in/mcp (streamable HTTP, no auth, no install)

LiquiLens is the institution-evidence product of LIQUILENS PRIVATE LIMITED, a registered company in India (CIN U62011RJ2026PTC116792). One company, three connected products: LiquiLens investigates institutions, Seiche explains funding conditions, and Undertow examines market liquidity. People and AI agents can follow an exposure across all three while retaining each product's source dates, coverage gaps and research limits. International access does not imply universal market coverage or a shared score.

Read the company and investor story and the shared company profile.

Try it live: liquilens.in/developers · REST catalog: api.liquilens.in/api

LiquiLens is a failure early-warning system for banks and lenders, built on public data with a machine-readable historical-evidence boundary served beside every claim. Anonymous discovery on 11 October 2026 returned MCP 1.8.1 with 24 read-only tools and 5 guided prompts, so an agent reads the same status, eligibility flags and cited record a human sees. The repository contract separately pins MCP 1.8.1 at d6f99bdf2de14e8ef2f29032cf9561ca8a446de8, with 23 read-only tools and 5 guided prompts. The capability tables below describe that pinned source. Discover the hosted endpoint for the current public inventory, including institution_evidence_monitor.

It now reads both ends of the chain: which institutions are fragile, and whether that stress is actually crossing into the real economy: companies rolling paper and drawing credit lines, households falling behind. Channels that cannot be read are named in cannot_see rather than reported as calm.

Add it

For Hermes and OpenClaw, use the native setup guides:

The current guides select thirteen research tools across LiquiLens, Seiche and Undertow, including GIFT City, reference FX and gold scenarios. Their dated native-client receipts retain the original nine-tool selection and exact verification scope. Public research requires no account or API key; fair-use limits apply and your model provider may charge separately.

The free starter kit also includes a Python brief, a manual n8n funding workflow and configurations for the clients below.

OpenClaw users can also install the optional research instructions from ClawHub:

Connect the MCP servers using the OpenClaw guide first. The skill supplies task instructions and a configuration template; installation does not configure servers or run research. Its source bundle is MIT-0; that license does not relicense source data or API responses.

Checked on 4 October 2026 (UTC): version 1.1.0 is public and ClawHub's version-pinned verification result reports a passed security check with a benign, high-confidence verdict. Its embedded detailed scanner report retains seven findings, including medium external-transmission flags for the public MCP URLs, and reports partial analysis. Review the versioned security audit before installation and use public research inputs. Its isolated Linux install recorded the correct version and matched all three reviewed source files. ClawHub's public verifier and the installed Linux client's version-pinned openclaw skills verify now both report pass, with the generated Skill Card present. The upload is unsigned and has no server-resolved GitHub import provenance. A fresh isolated macOS install after storage recovery also matched all three source-file hashes, and its version-pinned verification reports pass. The earlier macOS timeout remains a separate failed attempt. No model or research tool ran in these checks. The direct MCP configuration is available independently of the ClawHub skill.

Claude Code:

Codex:

Cursor: merge the remote MCP config into .cursor/mcp.json for one project or ~/.cursor/mcp.json for all projects. The download includes the companion Seiche funding endpoint. Preserve existing entries. For an existing Codex config, use the TOML snippet and merge its server entries into ~/.codex/config.toml.

Claude.ai / ChatGPT: add a custom connector or MCP app with the endpoint above where that feature is available in your workspace.

This repository is the public discovery and documentation mirror. The hosted implementation is maintained in a private core repository. This public mirror pins its inspectable capability contract to the reviewed source revision above. The official Registry entry uses the name io.github.beepboop2025/liquilens. This repository's Registry metadata revision 1.8.1 points to the public mirror and free starter kit. The hosted MCP now reports version 1.8.1 at the same endpoint. The Registry metadata and runtime happen to share a version number; they remain separately verified contracts. The pinned source and tool inventory above describe the current runtime. The existing Registry record continues to identify this public mirror.

The current runtime separates fresh model results from reviewed filing facts: expired filing scores are excluded from the live board, current disclosures retain their official source and publication clocks, and inactive institutions remain identifiable in historical evidence. Its US evidence includes the June 2026 panel and reviewed 2026 failure notices through 25 September.

Query source histories

The separate source-data MCP and Python client expose funding histories, bank filings and Bitcoin/Liquid settlement observations with source receipts, native units and capture clocks. Use the live coverage table or import its OpenAPI contract.

Run a research task

The research recipes run with Python 3.11+ and its standard library. No API key, LLM or Python package install is needed. Download and inspect financial_research.py, then run:

python3 financial_research.py bank-reviewpython3 financial_research.py bank-review --slug cosmos-ucbpython3 financial_research.py funding-brief

python3 financial_research.py bank-reviewpython3 financial_research.py bank-review --slug cosmos-ucbpython3 financial_research.py funding-brief

The first command discovers covered bank slugs. The second checks coverage before retrieving that exact bank's sourced asset-quality history; an absent slug stays not_covered. The funding brief calls Seiche's money-market desk. Returned evidence retains its native source dates, eligibility, unavailable states and limitations; a successful request does not imply fresh or complete evidence. Each run is bounded and performs no scheduling or trading. Operators should add --verification so their checks are labelled synthetic and excluded from adoption totals.

For a visual workflow, download the n8n bank-review workflow and follow the import guide. It uses a manual trigger and no LLM. The guide records its execution-verification status; a downloadable workflow does not imply acceptance into n8n's template library.

Protocol compatibility

2026-07-28: stateless requests use server/discover, per-request _meta, MCP-Protocol-Version, and mirrored Mcp-Method / Mcp-Name routing headers.
2025-11-25, 2025-06-18, and 2025-03-26: retained legacy initialization, tools, prompts, notifications, batching, and ping behavior.
Tools, prompts, and discovery responses are deterministic. Modern list responses carry public cache metadata.
resources/list and resources/templates/list are supported and currently return empty catalogs. resources/read returns an explicit not-found error instead of inventing a resource.

Tools

Tool	What it serves
`bank_asset_quality_review`	Exact-slug bank review with sourced GNPA/NNPA history, percentage-point changes, distinct PCR definitions and capital/supervisory scope; no new score or credit approval
`bank_npa_reconciliation`	Arithmetic check of a complete caller-supplied NPA movement table; keeps cash recoveries, write-offs, sales and upgrades distinct without authenticating the input
`banking_specialisation_coverage`	Discover covered Indian commercial, small finance and urban cooperative banks, with observed, stale, historical and absent evidence distinguished; not a census or rating
`corporate_transmission_board`	Is funding stress reaching nonfinancial firms? The commercial-paper market, bank credit lines, real-economy confirmation (claims, capex, inventories, openings, business bankruptcies) and a balance-sheet context channel, with a TRANSMITTING/CONTAINED verdict
`crypto_exposure_board`	Cited bank/crypto exposure register joined to Undertow run-risk context; display-only, never an institution score
`crypto_regime_board`	Compact BTC/ETH change-point state and its display-only cross-read against disclosed bank exposure
`evidence_europe`	`NAMED_CASE_FILES_CONSTRUCTION_PIT`: seven audited case files, deliberately no cohort claim; both eligibility flags false
`evidence_india`	`PERIOD_END_PROXY_CONSTRUCTION_PIT`: 48 institutions across two decades, misses and false alarms included; both eligibility flags false
`evidence_institution`	One Indian institution's construction-PIT crisis replay with sourced quarterly rows
`evidence_markets`	Historical-evidence status for all three markets, with `validated_backtest_eligible` and `real_money_eligible` served explicitly
`evidence_us`	`CURRENT_AMENDED_CONSTRUCTION_PIT`: 552 FDIC failures since 2008, 72.8% recall, 21.7-month median lead and AUC 0.854; both eligibility flags false
`failure_radar_board`	The live board: every Indian lender with a fresh vetted dossier, including failure PD term structure (12/24/36m), RBI action-zone status, funding fragility, market distance to default, and watchlist tier under a published rule
`failure_radar_institution`	One institution in depth: PD trajectory with drivers, PCA/SAF headroom history, forensic screen, market reading
`forward_odds`	Counted forward stress odds for each public-signal layer, withheld until the layer has enough observed history
`household_credit_board`	Is stress transmitting through household balance sheets? Fed delinquency and charge-off legs against each leg's own trailing decade, two-sided revolving-credit velocity, debt service as unscored context
`institution_research_coverage`	Discover institution dossiers, registry entries and reviewed filing facts with separate observed, stale, historical, subset and registry-only states; coverage is not a rating or permission to trade
`institution_review_packet`	One lender's deterministic evidence packet for human review, with coverage and freshness stated explicitly
`latest_article`	Today's exact evidence-led LiquiLens article, or a labelled historical replay when the evidence did not move
`rbi_supervisory_tape`	Latest RBI enforcement actions, each linking to the RBI's own page
`research_network`	Bounded source discovery for connected Palimpsest, Seiche and Undertow research with availability and evidence boundaries intact
`stablecoin_rails_board`	Issuer peg, redemption-run, chain-concentration and rail tripwire state; missing data never becomes `CALM`
`universe_search`	RBI's official registered-NBFC registry (9,000+ entries)
`verify_published_record`	Independent cryptographic verification of the as-published record

Prompts

Prompt	Guided playbook
`bank_asset_quality_brief`	Discover an exact bank slug, review cited NPA history and capital scope, then reconcile only a complete disclosed movement table; preserve stale/missing evidence and supervisory gaps
`crypto_liquidity_briefing`	BTC/ETH regime, stablecoin rails, and disclosed bank links in one display-only evidence pack
`failure_radar_briefing`	Board-level institution risk with transmission context from the public US signal layers
`institution_health_check`	One lender health check with the historical-evidence boundary and uncertainty attached
`stress_evidence_pack`	Region-specific evidence for India, the United States, Europe, or the cross-market summary

The governance line

The generative layer explains; deterministic screening policy scores. Historical diagnostics retain their served evidence status and eligibility flags and do not become validated-backtest or real-money evidence merely because a deterministic engine produced them. Screens, not ratings; not investment advice. Institutions without a vetted dossier are absent by design — the tools say so rather than inventing a score.

The three historical status tokens are part of the API contract, not marketing copy: PERIOD_END_PROXY_CONSTRUCTION_PIT for India, CURRENT_AMENDED_CONSTRUCTION_PIT for the United States, and NAMED_CASE_FILES_CONSTRUCTION_PIT for Europe. In the current release, every market serves validated_backtest_eligible: false and real_money_eligible: false.

Siblings from the same lab

Seiche — US money-market funding stress (the plumbing)
groundcheck — claim grounding and citation verification
Palimpsest (https://api.seiche.info/palimpsest/mcp) — live internet-censorship signals
Undertow — the cross market liquidity map: daily tiered board, exit cost at position size, Telegram front door at t.me/undertow_LiquiLens_bot

Product: liquilens.in · Live demo: demo.liquilens.in

Directory listing

[AllMCPs listing]

This account-linked marker supports ownership review of the existing directory record. Directory badges and cached tool counts are separate from the live MCP contract and its evidence limits.

Quant research agent integrations

Native framework tools and cited quant pipeline captures connect Seiche funding, LiquiLens bank diagnostics and Undertow market liquidity through compact read-only tables. LangChain/LangGraph, CrewAI, OpenAI Agents and Pydantic AI share one evidence contract and repeat-call revision tokens. Current published history is not an as-published vintage archive. No execution authority or institutional-adoption claim is implied.

来源：README.md，提交 0e1ee5e

工具

research_workspace

Read the same coverage, source freshness, gaps and descriptive overview as the website. Compare snapshot_id across REST/MCP; independent source clocks remain explicit.

输入参数product

research_series_summary

Analyze the latest captured series for one exact entity, with adjacent change, trailing-window missingness, native units and source receipts. Current analytics are withheld for stale, unavailable or unverified inputs.

输入参数entityseries必填windowdataset必填

research_tracked_institutions

Search the existing Indian dossier and RBI registry populations, retaining financial-evidence gaps and source clocks. Populations overlap and are not a global census.

输入参数qlimitscopeoffset