What is new

Recent releases, migrations and guides across the MCP servers, newest first.

• QDESK: a reporting desk for portfolio reports and financial dataSep 2026. Ask in plain language and it looks up the symbol, then runs the report. Also reachable on API and MCP if you want it wired into your own tools.
• IPL ball by ball is live on the cricket API, download and MCPSep 2026. 1,200+ matches since 2008, alongside Test, ODI and T20 internationals for men and women back to 2001.
• Markdown to PDF rebuilt, and your agent can drive it over MCPAug 2026. Headings, nested lists and code blocks all survive the conversion now. Paste the text yourself or hand the job to an agent.
• A database server with API and MCP you can point an AI agent atAug 2026. Read-only SQL over two engines, Postgres and DuckDB, through a REST API and an MCP connector. No key, no auth.
• Technical analysis on any ticker without visiting the siteAug 2026. Charts, indicator readings and an outlook for any Yahoo Finance ticker, reached entirely from an agent.
• Four routes into QuantStats, depending on what your AI can doAug 2026. Which route fits a chat, a coding agent or an MCP client, and why a plain chat cannot do it alone.
• Every Tigzig MCP is now Streamable HTTPJul 2026. ACTION FOR EXISTING USERS: every server now lives at its host plus /mcp. If a Claude, ChatGPT or LM Studio config still points at the old SSE endpoint, change it once and it keeps working.
• Yahoo Finance MCP backend upgraded: 25 tickers in one requestJul 2026. Batch calls take up to 25 tickers, answers come back in seconds, and a bad ticker returns a short note instead of stalling the whole request.

The MCP servers

Seven servers, all open and no-auth, over Streamable HTTP. Each links to its docs page with the endpoint, transport and tool list. The same backends also serve plain HTTP APIs.

• TREMOR - Macro & Credit IndicatorsAbout 330 curated macro and credit-stress indicators (FRED, OECD, RBI, NY Fed, NCUA, FDIC, Shiller, NSE). The same data is an open HTTP API too.
• Database Query - read-only SQLRead-only SQL against Postgres and DuckDB, with a multi-layer security stack. Reach it through a connected GPT or any MCP client.
• Compare Securities (QRep)Compare up to 6 securities across 80+ QuantStats metrics - returns, drawdown, Sharpe - as PDF/HTML reports.
• Technical AnalysisA technical-analysis report for any ticker: indicator computations plus a Gemini-Vision chart read. PDF and web output.
• Security Performance Report (FFN)A multi-security performance and risk comparison (returns, drawdown, Sharpe) from a list of tickers.
• Yahoo Finance ExtractorPrice, profile, financials, analyst targets, estimates and holders for any ticker (about 20 endpoints). No key.
• Markdown to PDFConvert Markdown to a formatted PDF as a tool call (ReportLab). The PDF engine behind several Tigzig tools.

Connect & use MCP (guides)

Hands-on guides for wiring Claude, ChatGPT and Excel to these live backends over MCP. Start with the setup guide.

• MCP Guide for Claude in Excel - and any MCP clientThe setup guide. Start here to connect an MCP client.
• Talk to Your Database from Excel - Postgres, DuckDB, via Claude + MCPQuery a live database in plain English from the sheet.
• Talk to Your Database from Excel - Part 2Deeper patterns.
• Claude in Excel with MCP Connector - talk to your backendsReach any backend from inside Excel.
• Pull your quant numbers with Claude / ChatGPT in Excel (QRep MCP)81 risk/return metrics into the sheet over MCP.
• Claude in Excel + MCP + xlwings Lite + Claude Code - four tools togetherPull data, build a model, validate it, end to end.

Database MCP in practice

The Database Query server is the one that runs caller-supplied SQL, so it is the one with the most written about it. Ball-by-ball cricket is the open dataset behind it. These cover what the server does, how callers reach it, and what goes wrong.

• A database server with API and MCP you can point an AI agent atStart here. Read-only SQL over Postgres and DuckDB through both a REST API and an MCP connector, free and keyless.
• Run your own SQL on my database. Free, no signup and no keyWorked queries against more than two million deliveries, from either the API or an MCP client.
• The commonest failure is the URL, not your SQLPercent encoding is where a GET query goes wrong, in a few predictable ways. Read this before filing a bug.
• Both formats on both engines: run your SQL on eitherA query no longer has to follow the data to a particular engine. What that took, and why it matters to a caller.

Build your own MCP server

How these servers are built - FastAPI wrapped as MCP, and mounting them in low-code agent flows.

• ChatGPT connected to integrated FastAPI-MCP serversThe FastAPI-as-MCP pattern, with a technical-analysis report as the worked example.
• Build AI workflows with MCP servers + n8nMount an MCP server in an n8n flow.
• My public MCP server got hammered - security lessons from a bot attackWhat breaks when you expose an MCP to the internet, and how to harden it.

Live apps & GPTs on the same backends

Want to see a backend in action? These run on the same MCP servers and APIs above.

• QDesk / Portfolio Analysis SuiteThe Technical Analysis, Security Performance and Yahoo Finance backends together - QuantStats, SPR and TA reports in one place.
• Database AI hubChat with any database. Ask for a PDF and it calls the Markdown-to-PDF backend.
• Markdown to PDF - web toolThe web front end for the Markdown-to-PDF backend. Paste Markdown, get a typeset PDF.
• Custom GPTsChatGPT custom GPTs wired to these same backends - Multi-Database, Quants and Technical Analysis.
• Flowise / n8n agent flowsLow-code flows that mount these MCP servers to produce reports and PDFs.

MCP security

An MCP server is an API surface like any other, so the same hardening applies - how much depends on what your server actually does. A data-only server needs rate limiting and input validation; a SQL server needs the full database hardening (read-only enforcement, query validation, resource-exhaustion blocks); one that wraps a backend adds the FastAPI checks.

• Four days of an open SQL endpoint, and the SSRF someone found in the first hoursAug 2026. What real callers were refused on a keyless SQL endpoint, and what those refusals showed about how it gets probed.
• Security checklist - MCP section (and full 120-item list)React, FastAPI, Postgres, DuckDB, Cloudflare and a dedicated MCP section. Each item: risk + solution + code.
• Security lessons from a real MCP bot attackA live incident write-up.

For AI agents - the machine directory

For a machine-readable list of every server with its endpoint and transport, point your client at the directory - each server self-describes via MCP initialize.

• MCP directory (servers.json, SEP-1649)The machine directory of every MCP server.
• Site index (llms.txt)The whole-site directory for AI agents.
• API catalog (RFC 9727)Every HTTP endpoint incl. full-dataset bulk downloads.
• Point your agent at api.tigzig.com and it works out which dataset it needsAug 2026. One address, close to 38 million rows across India and the US, and the agent pulls the file itself.
• Tigzig is built AI-agent first: the fastest way in is to ask your AIAug 2026. Worked example prompts for ChatGPT, Claude, Gemini and Meta AI.