lk-tax-mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@lk-tax-mcpHow much APIT will be deducted from my Rs. 350,000 monthly salary this month?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
lk-tax-mcp
Exact, sourced Sri Lankan tax calculations for AI assistants. An MCP server that lets Claude (or any MCP client) calculate APIT, income tax, withholding tax, VAT and SSCL from the official Inland Revenue Department rules for any date since 1 January 2023, instead of recalling rates from memory.
What it looks like
A real Claude Code session with only this server connected:
You: I'm paid Rs. 350,000 a month in Sri Lanka. How much APIT will my employer deduct this month, and how much was it before the April 2025 change?
Claude calls calculate_salary_tax twice, once for today and once for 31 March 2025, and answers (abridged):
This month: Rs. 32,500 APIT. The first Rs. 150,000/month is tax-free, then 6% / 18% / 24% / 30%. Source: IRD Notice PN/IT/2025-01 and APIT Tax Table No. 01 (2025/26). In force since 1 April 2025.
Before the change (to 31 March 2025): Rs. 52,500. Rs. 100,000/month tax-free, then 6% / 12% / 18% / 24% / 30% / 36%. Source: APIT Tax Table No. 01 (2023/24 & 2024/25).
So you now pay Rs. 20,000 less per month. This assumes it's your regular pay from your main job and you're a Sri Lankan tax resident; one-off bonuses or a second job use different tables.
The numbers come from the tool, the caveats come from the server's instructions, and every figure is traceable to an IRD document.
Related MCP server: JP Invoice & Tax MCP
Why
Language models are confident about tax rates, and Sri Lanka's change often: in 2025 the salary tax bands were rebuilt, and in 2026 alone the SSCL threshold dropped, VAT on financial services rose to 20.5% and withholding tax was extended to more professions, all mid-year. A model recalling "the rate" is wrong for some dates and has no source to show. This server gives the model exact arithmetic over dated, sourced rules, built on lk-tax, which is tested against every row of the IRD's official APIT tables.
Install
Requires uv (uvx runs the server without a manual install).
Claude Code
claude mcp add lk-tax -- uvx lk-tax-mcpClaude Desktop: add to claude_desktop_config.json (Settings → Developer → Edit Config):
{
"mcpServers": {
"lk-tax": { "command": "uvx", "args": ["lk-tax-mcp"] }
}
}Any other MCP client: run uvx lk-tax-mcp as a stdio server, or pip install lk-tax-mcp and run lk-tax-mcp.
Tools
Tool | Answers |
| APIT to deduct from a month's salary (official Tax Table No. 01 amount) |
| Income tax for a year of assessment, with band breakdown |
| WHT on 11 payment types: thresholds, exemptions, grossing up |
| Add VAT to a price or extract it from a VAT-inclusive price |
| Does a business need to register for VAT? |
| SSCL on turnover by activity (liable share × 2.5%) |
| Does a business need to register for SSCL? |
| Rates, bands and thresholds in force on a date |
| Every change to a tax since 2023, with dates and sources |
Every result includes a one-line summary, exact amounts as decimal strings, and the rule version used
(effective_from, effective_to, IRD sources).
Design notes
Read-only and offline. Every tool is a pure calculation, annotated
readOnlyHintandidempotentHint. No network access, no files, no credentials.Dates are explicit. Each calculation takes the date it applies to (pay date, invoice date), so "what was it last year?" gets last year's rules.
Errors the model can act on. Bad input comes back as a tool error explaining what to fix (
monthly_income must be a number,lk-tax covers 2023-01-01 onwards), not a generic failure.Choices are enumerated. Payment types and SSCL activities are enums in the schema, each described, so the model picks a valid one.
Honest about gaps. The server instructions tell the model what isn't covered (lump sums, non-resident employees, VAT exemption classification, dates before 2023) and to say so rather than estimate.
Development
uv sync
uv run pytest # includes a test that launches the real command over stdio
uv run mcp dev src/lk_tax_mcp/server.py # explore the tools in the MCP InspectorBuilt with the MCP Python SDK v2.
Disclaimer
Not tax advice. Results link to the IRD sources they come from, so you can check them.
License
MIT
This server cannot be deployed
Maintenance
Related MCP Connectors
Source-backed Sri Lanka payslip tools for APIT/PAYE, EPF, ETF, stamp duty, and net pay.
Mobile money fees, African tax rates, currency conversion and Uganda PAYE for AI agents.
Nigeria PAYE, VAT and CBN rates, sourced and dated. Plus invoice totals, QR codes, JSON, Base64.
Jordan & GCC labour-law, tax and finance calculators for AI agents — 36 deterministic tools.
Related MCP Servers
AlicenseAqualityAmaintenanceOpen-source, accountant-verified tax computation skills for AI agents. 261+ skills across 172+ jurisdictions covering income tax, VAT/GST, payroll, corporate tax, crypto, and cross-border planning. Every skill is verified section-by-section by licensed CPAs and chartered accountants. 3 tools (list_skills, get_skill, get_skill_sections) and 1 prompt (skill-review).123432AGPL 3.0- AlicenseNot gradedqualityCmaintenanceProvides Japanese tax and invoice utilities such as consumption tax calculation, withholding tax, invoice number validation, and tax rate summarization, enabling AI assistants to perform these operations locally without external APIs.MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to perform Japanese invoice and tax calculations, including consumption tax, withholding tax, invoice number validation, and invoice data generation, all locally without external APIs.MIT
- AlicenseAqualityCmaintenanceEnables AI assistants to calculate Swiss income, wealth, inheritance, capital payment, and corporate taxes for any municipality, compare tax burdens, and find cheapest locations without an API key.13MIT