Tijori Finance README

Tijori Finance MCP

India's first MCP server for Indian equity research.

Talk to 5,000+ NSE/BSE listed companies directly from Claude with India's equity research MCP

What you can do

Code

"Pull Zomato's last 6 quarters of revenue and EBITDA margin, then read their
latest concall transcript — are they hitting the targets management guided for?"

"Bajaj Finance's FII stake dropped 400bps last year while promoters held flat —
pull their latest investor presentation to figure out if this is distribution or passive rotation."

"Crude is up — check the raw materials data for chemical spreads, then pull
revenue mix and operational metrics for Deepak Nitrite to see if margins are at risk."

"The goal is to turn data into information, and information into insight." — Carly Fiorina

In Action

1 — Cross-tool deep dive: Titan

Revenue mix pulled first, then cross-referenced against P&L data to explain trends

Titan — Revenue Mix × P&L

2 — IndiGo: P&L history

Full income statement history retrieved in one call — revenue, EBITDA, PAT across years

IndiGo — P&L History

3 — Concall → Financial projection

Knowledge base fetched the latest earnings call transcript, Claude read it and built a forward projection model

Concall → Financial Projection

Why this exists

Most financial MCP servers are built for US markets — Yahoo Finance, SEC filings, S&P data.

Nothing existed for India.

Tijori Finance is the most comprehensive source for Indian equity data — operational KPIs, revenue segment breakdowns, market share trends, and curated investor documents that aren't available anywhere else. This MCP server exposes all of it to Claude.

Feature	Tijori Finance MCP	Traditional Research	Bloomberg / Refinitiv
Setup Time	5 minutes	Hours (Python, Excel...)	Weeks (Contracts)
Cost	Free + Tijori subscription	Variable	$30k+/year
Indian Market Coverage	✅ 5,000+ NSE/BSE stocks	❌ Fragmented / manual	✅
Operational KPIs	✅ Segment data, KPIs, market share	❌ Manual scraping	✅ Proprietary
Concall Transcripts	✅ Earnings calls, investor docs	❌ PDF hunting	Partial
AI-Ready Output	✅ Structured JSON → Claude	❌ Unstructured pages	❌ Proprietary only
API Keys Required	None	Multiple (OpenAI, etc.)	N/A

What makes this different from a web search?

A web search gives you unstructured pages. This gives Claude structured, queryable data — historical time series, typed fields, and consistent schemas across all 5,000+ companies. Claude can reason over it, compare across companies, and build analyses — not just summarize a webpage.

Tools

v1 — Stable

Tool	What it does
`search_company`	Search any company by name, returns slug
`get_company_overview`	Key ratios, forensics score, market cap, PE, ROE, ROCE
`get_financials`	P&L, Balance Sheet, Cash Flow, Ratios, Quarterly results
`get_shareholding`	10-quarter promoter / FII / DII / public breakdown
`get_operational_metrics`	All operational KPIs with full historical time series
`get_fund_flow`	Capital allocation breakdown over 1/3/5/7/10 years
`get_revenue_mix`	Segment breakdown with historical trend per segment
`get_market_share`	Market share % per metric with as-of date
`get_knowledge_base`	Annual reports, earnings releases, investor presentations, conference calls — returns URLs grouped by type
`fetch_document`	Fetch and extract full text from a PDF URL returned by `get_knowledge_base`. Reads through the authenticated session to bypass CDN access controls
`get_raw_materials`	Commodity price performance — chemicals, spreads, metals
`get_macro_indicators`	India macro — credit, IIP, GST, auto sales, GDP, trade
`get_markets`	Index performance — Nifty, sector indices, conglomerates
`get_sector_constituents`	All stocks inside a TJI niche sector index. Pass `tjiid` from `get_markets("niche")`. Returns slug, market-cap weight, and 1D–10Y price returns per stock
`get_conglomerate_constituents`	All companies inside a business group. Pass `tjiid` from `get_markets("conglomerates")`. Returns slug and 1D–10Y price returns per stock
`resolve_company_ids`	Resolve slug to numeric company ID

v2 — Stable

Tool	What it does
`list_popular_screens`	Browse Tijori's pre-built stock screens, grouped by category, with each screen's description and underlying query
`screen_companies`	Run a popular screen by name (`preset`) or screen 5,000+ stocks with Tijori's full query language — `%` values, field-vs-field comparisons, arithmetic like `capex/Net Block > 0.5`, plus business-data queries (`alternate`) like `market share > 50` or `revenue from Defence > 50`
`search_screener_fields`	Search the ~1,500-metric field catalog (with time-period variants like `3yr Avg ROCE`, `10Yrs ago PAT`) to find exact field names for `screen_companies`

What's New

2026-06-12 — Screener overhaul

Alternate data queries. The Advanced Screener's second box now works ad-hoc: screen_companies({ alternate: "market share > 50" }) or { alternate: "revenue from Defence > 50", filters: "Market Capitalization > 1000" }. Relationships: makes, revenue from, market share, uses, caters to, has plant in, combinable with AND/OR/NOT.
The three screener checkboxes. latest_results_only (only companies that reported the latest quarter), superstar_investors (only companies a whale investor holds, with the holder named per row), and sme (search the SME-listed universe).
Popular screens actually run now. list_popular_screens previously returned queries with the AND glued to the next field name (newlines in the page's links were being stripped), and dropped each screen's category, description, market-share query, and superstar-investor flag. It now returns all of those, and screen_companies { preset: "Monopoly Companies" } runs any of them through the same endpoint the website uses — including the ones advanced queries can't express.
The advanced screener exposes Tijori's full query language. Queries support every field in Tijori's catalog, % values (ROCE > 20%), comparisons between fields (Net Sales > 3Yrs ago Net Sales), and arithmetic (capex/Net Block > 0.5).
New search_screener_fields tool. Tijori has ~1,500 financial metrics once you count time-period variants (3yr Avg ROCE, 5yr Growth Net Sales, 10Yrs ago PAT…). Search the catalog to find the exact name, then use it in a query.
Result caps + pagination. Screens used to return every matching row (1,000+ for loose queries). Results are now capped at 50 by default (limit raises it) and offset pages through the rest — "show me the next 50" is served instantly from cache without re-querying Tijori. total_results always reports the full count.

2026-06-10 — Reliability & performance

Long, multi-tool queries no longer time out. Asking for a full workup in one go — e.g. "build a model for MTAR: overview + financials + concall + revenue mix + operational KPIs" — used to overwhelm the browser and fail with page.goto: Timeout errors. Page loads are now faster (ads, trackers and images are skipped), capped to a safe number of concurrent loads, and retried once on a transient timeout. Tools also wait for the actual data to render rather than a fixed delay, so you get complete results instead of half-loaded ones.
Fixed a startup browser leak. A burst of parallel tool calls on a cold start could spin up several Chromium instances and leave most of them orphaned. It now launches exactly one and shares it.
get_market_share returns cleanly when there's no data. For companies that simply have no market-share metrics (most non-lenders), it now returns an empty result with a note instead of erroring — which previously made the assistant retry the same call in a loop.

Setup

Requirements

Node.js v18+ — install the LTS version
A Tijori Finance account — free account works but has data limits; Pro recommended for full access
Claude Desktop

Option 1 — Wizard (recommended)

The setup script handles everything automatically: installs packages, downloads the browser, authenticates, and writes the Claude Desktop config for you.

Windows

Download Node.js and install it (choose LTS)
Download Claude Desktop and install it
Download this repo and unzip it anywhere
Double-click setup.bat inside the folder
Follow the prompts — enter your Tijori email/password, then log in through the browser window that opens
Fully quit and reopen Claude Desktop

Mac

Download Node.js and install it (choose LTS)
Download Claude Desktop and install it
Download this repo and unzip it anywhere
Double-click setup.command inside the folder — if macOS blocks it ("unidentified developer"), right-click the file and choose Open the first time
Follow the prompts — enter your Tijori email/password, then log in through the browser window that opens
Fully quit and reopen Claude Desktop (Cmd+Q, not just closing the window)

Linux (or any terminal)

bash

git clone https://github.com/LaZZy0v0/tijori-finance-mcp.git
cd tijori-finance-mcp
node setup.js

Follow the prompts, then fully quit and reopen Claude Desktop.

Option 2 — Manual

For those who want to see exactly what each step does.

1. Clone and set credentials

bash

# Mac / Linux
git clone https://github.com/LaZZy0v0/tijori-finance-mcp.git
cd tijori-finance-mcp
cp .env.example .env

# Windows
git clone https://github.com/LaZZy0v0/tijori-finance-mcp.git
cd tijori-finance-mcp
copy .env.example .env

Open .env in any text editor and fill in your Tijori credentials:

env

TIJORI_EMAIL=your@email.com
TIJORI_PASSWORD=yourpassword

2. Install packages and browser

Terminal

npm install
npx playwright install chromium

npm install downloads the Node.js dependencies. playwright install chromium downloads a ~150 MB Chromium browser used to maintain your Tijori session — one-time only.

3. Authenticate

bash

node discover.js

A browser window opens at the Tijori Finance sign-in page. Log in as you normally would. Once you're in, the script visits a few pages in the background to capture API endpoints, then the window closes automatically. Your session is saved to output/session.json.

4. Configure Claude Desktop

Find your config file and add the block below. The path to src/index.js must be absolute.

OS	Config file location
Windows	`%APPDATA%\Claude\claude_desktop_config.json`
Mac	`~/Library/Application Support/Claude/claude_desktop_config.json`

config.json

{
  "mcpServers": {
    "tijori-finance": {
      "command": "node",
      "args": ["C:/Users/yourname/tijori-finance-mcp/src/index.js"]
    }
  }
}

5. Fully quit and reopen Claude Desktop

Test it

"Search Tata Steel using Tijori MCP"

If Claude returns a result, you're connected.

Session expiry

Tijori sessions expire periodically. When tools stop working:

Terminal

npm run reauth

A browser window opens — log in manually and you're back.

How it works

The server uses Playwright to maintain an authenticated browser session with Tijori Finance. Each tool navigates to the relevant page or calls the underlying API, parses the response, and returns structured JSON to Claude. Results are cached in-memory (6 hours for financials, 30 minutes for metrics) to keep things fast.

Architecture

Disclaimer

This project is not affiliated with Tijori Finance. It requires your own paid Tijori Finance subscription. Use for personal research only — do not redistribute the underlying data.

Built with Model Context Protocol · Data from Tijori Finance