Supported Platforms & Client Compatibility

Verified implementations across leading AI assistants, developer IDEs, and local offline models.

Platform	Local stdio (Private)	Auto-Config Support	Network / LAN Support
Claude Desktop	✔️ stdio + .mcpb	✔️ 1-Click	✔️ Yes
Cursor IDE	✔️ stdio	✔️ 1-Click	✔️ Yes
Windsurf	✔️ stdio	✔️ 1-Click	✔️ Yes
Google Gemini / Antigravity	✔️ stdio	JSON snippet	✔️ Yes
Codex / ChatGPT Desktop	✔️ Plugins / stdio	Settings menu	✔️ Yes
Cline / Roo Code	✔️ stdio	JSON snippet	✔️ Yes
Ollama & LM Studio (Air-gapped)	✔️ 100% Offline	Offline profile	—

Diagnostic Tooling

CLI Diagnostics & Preflight Check (`tally-mcp doctor`)

TallyMCP includes a built-in diagnostic engine that validates your setup in seconds before you connect your AI assistant.

Run the 5-point diagnostic check in PowerShell:

tally-mcp doctor
# Or show discovered company names:
tally-mcp doctor --show-names

[✓]Tally Reachability: Tests TCP connect to port 9000 on loopback or configured LAN IP.

[✓]Company Detection: Verifies that a company book is open and responsive to XML queries.

[✓]Write Permission & Safety: Verifies local write-ahead log (WAL) directory permissions.

[✓]Client Registration: Inspects Claude Desktop, Cursor, and Windsurf config files for valid PATH.

Installation & Connectivity Setup

Connect your accounting records to your AI assistant in 3 simple steps.

Turn on Tally's HTTP Connector

In TallyPrime, press F1: Help ▸ Settings ▸ Connectivity ▸ Client/Server configuration.

TallyPrime — Client/Server configuration

TallyPrime acts asBoth

Enable ODBC ServerYes

Port9000

💡 Verification: Open http://localhost:9000 in your browser. You should see “TallyPrime Server is Running”.

Install TallyMCP on Windows

Run the Windows MSI installer. It registers tally-mcp on your PATH and creates a Desktop shortcut for the Companion UI:

Download Windows Installer (.msi)Download Standalone .exe

Run Automated Setup Wizard

Open PowerShell and run tally-mcp setup. It automatically locates your installed AI clients and injects the correct configuration:

tally-mcp setup
# Output:
# === tally-mcp Setup Wizard ===
# [✓] Claude Desktop automatically configured!
#     Config file: C:\Users\User\AppData\Roaming\Claude\claude_desktop_config.json
#     Executable:  C:\Users\User\AppData\Local\Programs\TallyMCP\tally-mcp.exe
# [✓] Desktop shortcut created

Client-by-Client Setup Guide

Detailed instructions for Claude Desktop, Cursor, Windsurf, Gemini, and Codex / ChatGPT.

Claude Desktop (Local stdio — Recommended)

100% Local & Private

Inside Claude Desktop, click File ▸ Settings (or Ctrl + ,), choose the Developer tab in the sidebar, and click Edit Config.
Direct path: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "tally": {
      "command": "tally-mcp",
      "args": ["serve"]
    }
  }
}

Cursor IDE

Add to .cursor/mcp.json in your project workspace:

{
  "mcpServers": {
    "tally": {
      "command": "tally-mcp",
      "args": ["serve"]
    }
  }
}

Windsurf IDE

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "tally": {
      "command": "tally-mcp",
      "args": ["serve"]
    }
  }
}

Google Gemini / Antigravity

Add to ~/.gemini/mcp_config.json:

{
  "mcpServers": {
    "tally": {
      "command": "tally-mcp",
      "args": ["serve"]
    }
  }
}

Codex / ChatGPT Desktop

In ChatGPT Desktop: Settings ▸ Integration ▸ Plugins ▸ Add MCP Server.

Name: tally
Command: tally-mcp
Arguments: serve

Complete MCP Tools Reference (12 Tools)

All tools operate with compile-time safety and deterministic validation.

Tool Name	Access	Description	Parameters
tally_status	read	Get Tally reachability, loaded companies, and circuit breaker health	none
tally_list_companies	read	List all loaded and available companies in TallyPrime	none
tally_set_active_company	session	Set the active company for read operations in this session	company
tally_list_ledgers	read	List ledger accounts for a company	company
tally_trial_balance	read	Export Trial Balance report from Tally	company, from_date, to_date
tally_day_book	read	Export Day Book transactions from Tally (deterministically formatted as JSON)	company, from_date, to_date
tally_draft_voucher	draft	Validate and stage a voucher draft for out-of-band approval. Writes NOTHING to Tally.	company, voucher_type, date, number, party_ledger, entries, narration
tally_preview_draft	read	Preview human-readable summary and exact XML envelope for a drafted voucher	draft_id
tally_commit_draft	write	Post an approved voucher draft to Tally. Requires prior approval token from physical user click.	draft_id
tally_discard_draft	draft	Discard a pending voucher draft	draft_id
tally_extract_document	read/ai	Extract structured invoice/receipt data from a document file using configured AI provider or offline model	file_path
tally_run_report	read	Run a named report from an active tool pack (e.g. outstanding_ageing)	report_name, parameters

Real-World Conversation Recipes

Sample multi-turn prompts and deterministic workflow patterns.

Recipe 1: Account Balances & Trial BalanceRead-Only Safe

“Check our cash balance, bank account, and show the top 5 sundry debtors in Tally.”

AI invokes: tally_list_companies → tally_set_active_company → tally_list_ledgers

Recipe 2: Day Book Transactions (Structured JSON)Deterministic JSON

“Show all sales vouchers entered yesterday and calculate total GST collected.”

AI invokes: tally_day_book (returns typed voucher objects with voucher_type, amount, and ledger_entries).

Recipe 3: Drafting Vouchers (Two-Phase Commit)Human Approval Required

“Draft a payment voucher of ₹12,500 to Electricity Expense from our HDFC bank account.”

TallyMCP stages the draft locally, validates double-entry balance (Debits + Credits == 0), and emits an approval URL (http://127.0.0.1:9190/approve?token=...). The AI presents the URL to you. You review the XML preview and click Approve to post the entry into Tally.

Connecting to Tally Across Your Network (Wi-Fi or LAN)

Use TallyMCP directly on the PC/laptop where Tally is installed (localhost), or from any PC/laptop connected to the same network (including Wi-Fi or office LAN):

Pass --host and --port in your client arguments or environment variable:

{
  "mcpServers": {
    "tally": {
      "command": "tally-mcp",
      "args": ["serve", "--host", "192.168.1.50", "--port", "9000"]
    }
  }
}

Or set environment variable: TALLY_HOST=192.168.1.50. TallyMCP automatically enables LAN mode and validates connection latency.

Data Hygiene & Clean Uninstallation

TallyMCP puts you in complete control of your local logs, cache, drafts, and credentials.

CLI Cleanup Commands:

# Inspect data inventory without deleting anything:
tally-mcp cleanup --dry-run

# Clean temporary logs and cache with automatic backup:
tally-mcp cleanup --backup C:\Backups\TallyMCP

# Complete clean uninstall (removes data, PATH entries, and shortcuts):
tally-mcp uninstall

Troubleshooting Matrix

Symptom	Resolution
“Tally connection refused on 127.0.0.1:9000”	Verify Tally is open with a company loaded. Check `F1 ▸ Settings ▸ Connectivity` has “Enable HTTP Server” set to Yes on port 9000. Run `tally-mcp doctor` to diagnose.
Empty report envelope / No records	TallyMCP v1.0.0+ automatically catches empty report envelopes and informs your agent deterministically instead of returning blank XML tags. Ensure the active company has transactions within the requested date period.
Claude Desktop doesn't show Tally tools	Fully quit Claude Desktop from the system tray icon, not just by closing the window. Then restart Claude. Run `tally-mcp setup` to ensure the executable path is verified.
Tally returns `<LINEERROR>`	TallyMCP intercepts Tally line errors (e.g. unknown ledger or duplicate voucher number) and translates them into clear, formatted error messages so the AI agent can adjust its arguments.

Up and running in ten minutes.