Tally Prime MCP Server

fork of: dhananjay1405/tally-mcp-server

Tally Prime MCP (Model Context Protocol) Server implementation to feed Tally Prime ERP data to popular LLM like Claude, ChatGPT supporting MCP client. This MCP Server helps expose functionalities of Tally to LLM directly.

Prerequisites

• Tally Prime (Silver / Gold)

• Node JS

Ensure below things are pre-installed and setup:

• Ensure to download & install Node JS from official website

• XML Port of Tally Prime must be enabled (F1 > Settings > Connectivity > Client/Server configuration) with below settings

TallyPrime acts as = Server
Port = 9000

Note: Kindly avoid using Educational version of Tally Prime, which has limitations of date range. It will result in invalid / partial data being fed to LLM, leading to highly degraded & incorrect responses.

Related MCP server: Tally Prime MCP Server

Download

Avoid cloning repository directly. Utility is available for download (with required dependencies) on below link https://excelkida.com/resource/tally-mcp-server-v7.6.zip

One-click installer extension for Claude Desktop https://excelkida.com/resource/tally-mcp-server-v7.6.mcpb

Last updated: version 7.6 [04-Sep-2026]

Refer docs/CHANGELOG.md for details

Build from source

This project uses pnpm as package manager. It is needed only if you clone the repository and build the code yourself. Users of the zip file or the extension can skip this section.

Install pnpm (Node JS 22 or higher is required)

npm install -g pnpm

Corepack is an alternative, which ships with Node JS

corepack enable pnpm

Install the dependencies and build the project

pnpm install
pnpm build

The compiled files are written to the dist folder. Start the web-server version with

pnpm start

Run the tests

pnpm test

Check the build without writing anything, which is what CI should run

pnpm check

What pnpm build does

pnpm build runs three steps in order.

node scripts/bench-cache-insert.mjs is separate from the build. It measures how long the result cache takes to store a report, one row at a time against in batches, on rows held in memory. It needs no Tally.

Both generated outputs are committed, so a plain tsc still works. Do not edit src/templates.generated.mts or the tools array in manifest.json by hand. Change the .njk file or the tool definition and run pnpm build.

Project layout

Supported Platform

Implementation was tested on below AI platform

Setup (Local)

This mode of setup is to be used when MCP Client (like Claude Desktop, Perplexity etc.) and Tally Prime both exists in local PC. MCP Client software itself runs the MCP Server internally in such scenario.

Simply download & extract zip file somewhere on the disk. Assuming that we downloaded & extracted zip file on below path (folder)

D:\Software\Tally MCP Server

A sample setup for few popular tools is demonstrated.

Claude Desktop

Desktop version of Claude AI supports loading of local MCP server. Ensure you have Pro / Team / Max / Enterprise subscription of Claude, which supports higher limit compared to Free. MCP makes multiple calls to Tally for validation and inference, which might exhaust free limits quickly. Download Claude Desktop from following link claude.ai/download

One-click installation (via Extension)

Go to menu > File > Settings

Extensions > Advance Settings

Click on install extension button

Browse the extension file (with file extension mcpb) download at the start

A dialog window will appear asking Do you want to install Tally Prime? click Install button, which would install the Tally MCP Server

Installation via Config file (via Developer menu)

Go to menu > File > Settings > Developer

This will open My Computer window. Right click and edit claude_desktop_config.json file (via Notepad) with as below JSON

{
  "mcpServers": {
	  "Tally Prime": {
		  "command": "node",
		  "args": ["D:\\Software\\Tally MCP Server\\dist\\index.mjs"]
	  }
  }
}

Note: single slash in folder path needs to be substituted with double slash

Save the file. Close Claude Desktop (menu > File > Exit) and again re-launch it.

Verify by clicking on Tools button and check if Tally Prime appears in the list (screenshot below)

Perplexity Desktop

Perplexity Desktop version for MacOS supports connecting to local MCP server. Configuration file (JSON format) is same as demonstrated for Claude Desktop. In absense of MacBook, documentation with screenshot could not be written. Kindly refer to below blog on perplexity website, which explains the steps.

Perplexity Desktop MCP Connectivity

Setup (Cloud)

This mode of setup is to be used, when using browser-based MCP client like ChatGPT, Claude AI, Copilot, OR mobile-based app for these LLM which cannot access Tally Prime running inside local PC. In this scenario, MCP Server needs to run as web-server, internally connected to Tally securely. Setup is quite complicated, and is covered in detail in docs folder of this project.

• Linux-based Server

• Windows Server (exploration in-progress)

Available Tools

This server currently exposes 30 MCP tools.

What a reporting tool returns

Every tool that caches a result answers with the same envelope.

A tool that fails returns the reason as an error. A request Tally refused is an error, not an empty result.

metadata-collection

Returns metadata for supported collections.

Input No input.

Output JSON array with objects containing:

1. collection

2. description

query-option-values

Returns predefined option values used by input fields.

Input

Output JSON array of option values for the selected option name.

metadata-fields

Returns field metadata for a selected collection.

Input

Output JSON array of field metadata containing field name, description (if any), and normalized datatype (string, number, date, boolean).

query-database

Runs SQL query on in-memory pglite tables previously created by reporting tools.

Input

Output Query result in tab-separated text format.

query-collection

Queries a Tally collection for selected fields and caches output in an in-memory table.

Input

Output JSON: { "tableID": "..." }

list-master

Fetches list of masters for validation and auto-completion.

Input

Output JSON: { "list": [ ... ] }

chart-of-accounts

Extracts Chart of Accounts (or Group hierarchy) useful for preparing Balance Sheet, Profit and Loss, Trial Balance

Input

Output JSON: { "tableID": "..." } with columns:

1. ledger_name

2. group_name

3. primary_group

4. bs_pl (boolean) [true = Profit & Loss / false = Balance Sheet]

5. dr_cr (boolean) [true = Debit / false = Credit]

6. affects_gross_profit (boolean) [true = Affects Gross Profit / false = Does not affect Gross Profit]

7. sort_position (number)

trial-balance

Fetches trial balance for period.

Input

Output JSON: { "tableID": "..." } with columns:

1. ledger_name

2. group_name

3. opening_balance (number) [negative = Debit / positive = Credit]

4. net_debit

5. net_credit

6. closing_balance (number) [negative = Debit / positive = Credit]

profit-loss

Fetches profit and loss data for period.

Input

Output JSON: { "tableID": "..." } with columns:

1. ledger_name

2. group_name

3. closing_balance (number) [negative = Debit / positive = Credit]

balance-sheet

Fetches balance sheet data for period.

Input

Output JSON: { "tableID": "..." } with columns:

1. ledger_name

2. group_name

3. closing_balance (number) [negative = Debit / positive = Credit]

stock-summary

Fetches stock item summary for period.

Input

Output JSON: { "tableID": "..." } with columns:

1. stock_item_name

2. stock_group_name

3. opening_quantity (number)

4. opening_value (number) [negative = Debit / positive = Credit]

5. inward_quantity (number)

6. inward_value (number)

7. outward_quantity (number)

8. outward_value (number)

9. closing_quantity (number)

10. closing_value (number) [negative = Debit / positive = Credit]

ledger-balance

Returns ledger closing balance as on date.

Input

Output JSON: { "amount": number } where negative = Debit and positive = Credit.

stock-item-balance

Returns stock item closing quantity as on date.

Input

Output JSON: { "quantity": number, "unit_of_measurement": string } when found.

bills-outstanding

Fetches receivable/payable bill-wise outstanding as on date.

Input

Output JSON: { "tableID": "..." } with columns:

1. bill_date

2. reference_number

3. outstanding_amount

4. party_name

5. overdue_days

ledger-account

Fetches ledger account statement for period.

Input

Output JSON: { "tableID": "..." } with columns:

1. guid

2. date

3. voucher_type

4. voucher_number

5. alternate_ledger

6. party_name

7. amount (number) [negative = Debit / positive = Credit]

8. narration

stock-item-account

Fetches stock item account statement for period.

Input

Output JSON: { "tableID": "..." } with columns:

1. date

2. voucher_type

3. voucher_number

4. party_ledger

5. quantity

6. amount (number) [negative = Debit / positive = Credit]

7. narration

8. tracking_number

9. voucher_category

daybook

Fetches the daybook, one row per voucher entered in the period.

Input

Output JSON: { "tableID": "...", "rowCount": n, "totalRowCount": n, "hasMore": bool, "nextOffset": n } with columns:

1. guid (pass this to voucher-get)

2. date

3. voucher_type

4. voucher_number

5. reference

6. party_name

7. amount (number) [total debit value of the voucher, always positive]

8. narration

9. is_cancelled

10. is_optional

voucher-get

Fetches one complete voucher, with every ledger line, bill reference and inventory line. A ledger statement shows only the part of a voucher that touches that one ledger, this tool shows the whole transaction.

Input

Output JSON returned inline, not as a cached table:

1. voucher - the header, same fields as one daybook row

2. ledger_entries - ledger_name, amount [negative = Debit / positive = Credit], is_debit, cost_centre

3. bill_allocations - ledger_name, bill_name, bill_type, amount

4. inventory_entries - stock_item_name, quantity, rate, amount, godown_name, tracking_number

ledger-create-update

Creates or updates one or more ledger.

Note: This tool has ability to modify existing ledger. Always backup your Company before instructing this tool.

Input

Master ledger object accepts following

Output JSON result returned by import operation (success/failure details).

voucher-create-update

Creates new transactions, or replaces existing ones.

Note: This tool has ability to modify existing vouchers. Always backup your Company before instructing this tool.

Every ledger, voucher type and stock item named is validated first, and every amount of a voucher must add up to zero. If any voucher of the batch fails validation, nothing is sent to Tally at all.

Input

Voucher object accepts following

Output JSON counters returned by the import, plus verified, the vouchers found again in the daybook.

voucher-cancel-delete

Cancels or deletes existing transactions. A cancelled voucher keeps its number and stays visible with no entries, a deleted voucher is removed.

Note: This cannot be undone. Always backup your Company before instructing this tool.

Input

Output JSON result returned by the import operation.

group-create-update

Creates or updates accounting groups.

Input

Output JSON result returned by the import operation.

stock-group-create-update

Creates or updates stock groups.

Input

Output JSON result returned by the import operation.

unit-create-update

Creates or updates units of measurement. A simple unit carries a symbol and a formal name, a compound unit needs baseUnit, additionalUnit and conversion together.

Input

Output JSON result returned by the import operation.

godown-create-update

Creates or updates godowns, warehouses or locations.

Input

Output JSON result returned by the import operation.

stock-item-create-update

Creates or updates stock items or products.

Input

Stock item object accepts following

Output JSON result returned by the import operation.

company-create

Creates a new company. This is not the same as set-company, which only selects a company that already exists.

Note: Tally must accept company creation over the XML interface. Read the returned counters and confirm with list-master.

Input

Output JSON result returned by the import operation.

delete-master

Delete one (or more) masters from Tally

Input

A voucher has no name, it is identified by its guid, so use voucher-cancel-delete for one.

Output JSON result returned by delete operation (count of deleted, skipped, etc).

set-company

Sets active company context in Tally Prime.

Input

Output JSON string: "OK" on success.

set-period

Sets active reporting period context in Tally Prime.

Input

Output JSON string: "OK" on success.

Environment Variables

End-users are free to hard-code few settings which needs to be applied

Several Tally users on one machine

A Windows Server runs one copy of Tally per signed in person, and each copy needs its own XML port. MULTI_USER=1 turns on the profile registry so one server process can serve all of them:

• each profile is reachable at /u/<id>/mcp and talks only to its own Tally port

• each profile has its own consent password and its own bearer token

• a token issued for one profile is refused on every other profile

• the write tools can be hidden per profile

• a setup page at /admin, reachable from the machine itself and guarded by an admin token, finds the running copies of Tally and makes the profiles

The packaged Windows service does all of this. See docs/server-setup-windows.md to install it, and windows/README.md to build the setup EXE.

Security of the remote deployment

This applies only to the remote (HTTP) setup. The Claude Desktop extension runs over stdio and is not affected.

Copy .env.example to .env and set a real PASSWORD. .env is ignored by git and must never be committed. The server refuses to start on a non-local MCP_DOMAIN while the password is still the default.

Access control works as follows.

• Every request to /mcp needs a bearer token, including GET (the notification stream) and DELETE (ending a session).

• A session belongs to the OAuth client that opened it. A token issued to one client cannot read or end another client's session.

• The Host header must match MCP_DOMAIN, a loopback name, or an entry in ALLOWED_HOSTS. This blocks DNS rebinding.

• Tokens are checked for expiry, refresh tokens rotate on use and are bound to the client they were issued to, and an authorization code can be spent once.

• Repeated wrong passwords from one address are refused for a while.

Run the server behind TLS. Tokens and the password are sent in plain text over a plain HTTP connection.

Contact

Project developed & maintained by: Dhananjay Gokhale

Email: info@excelkida.com Whatsapp: (+91) 90284-63366