Model Context Protocol · Live

The ERES MCP ServerPakistan's first headless ERP

Connect ERES to Claude, ChatGPT, Cursor — or any AI agent you already use — and work your books in plain language. Ask what the company is worth, post an expense, pull a ledger, send a statement on WhatsApp. Same data, same permissions, no screens.

Works with

Claude
ChatGPT
Cursor
Any MCP client
Claude·connected to ERES
live
What headless means

An ERP you talk to, not one you navigate

Every other ERP in the market gives you screens. ERES gives you screens and a protocol — so the same accounting engine can be driven by a person clicking, or by an agent reasoning.

Not a chatbot bolted on

This is not an assistant that guesses inside your ERP. It is your ERP exposed as tools, with the same validation, the same hierarchy rules, and the same posting logic the dashboard uses.

Bring your own agent

You are not locked into our AI. Use Claude, use ChatGPT, use whatever comes next — MCP is an open protocol, so ERES keeps working as the model landscape changes.

Outperform the interface

One question can cross modules that would take a person six screens: ledger, inventory, payroll and tax, joined and answered in a sentence.

Setup

Connected in under a minute

No middleware, no data warehouse, no export. One URL and one consent screen.

01

Add the connector

Paste the ERES MCP URL into ChatGPT, Claude, Cursor, or any MCP-capable client. The client discovers ERES automatically and registers itself — there is nothing to build.

https://app.eres.cloud/api/mcp
Add ERES to your agent
Adding the ERES connector inside ChatGPT
02

Sign in as yourself

You authenticate against ERES directly, in an ERES window. Your password never reaches the AI client — ERES issues it an encrypted token instead.

Sign in as yourself
Signing in to ERES from an AI client to authorize the MCP connection
03

Approve the scope

One consent screen shows exactly what the agent may read and write. The connection is pinned to that organization — the agent cannot switch companies afterwards.

One consent screen
The ERES consent screen granting an AI agent scoped access to a company
In practice

Real answers, rendered properly

ERES returns structured results your agent can display as tables, tiles and charts — not a paragraph of numbers it might mistype.

Your whole company, answered
ChatGPT showing an ERES company overview with cash, receivable, payable and working capital
Cash and bank, receivables, payables, stock value, expense split and working capital — from one question.
Ask in plain language
ChatGPT rendering an ERES employee directory with headcount and payroll totals
"List all employees" returns headcount, monthly payroll and the full team — rendered as a real component, not a wall of JSON.
Tool catalogue

Everything the agent can reach

Grouped the way ERES itself is grouped. Read tools answer questions; write tools prepare work that still goes through your approvals.

Orientation

Where the agent starts: what this company is, what modules it runs, and what it is allowed to ask next.

get_company_profileget_overviewget_directory

Accounts & parties

Open accounts inside the five-level ERES hierarchy, and create customers or suppliers with their receivable and payable ledgers wired up.

get_chart_of_accountsget_partiescreate_chart_of_accountcreate_party

Money & balances

Bank and cash positions, party ledgers and trial balance, read from posted entries rather than a nightly copy.

get_cash_and_bankget_ledgerget_trial_balance

Financial statements

Balance sheet, income statement and daily movement — the same numbers the dashboard renders.

get_balance_sheetget_income_statementget_daily_report

Vouchers

A voucher type carries its own linked account and side, so the agent sends one half of the entry and ERES generates the contra.

get_voucher_typescreate_voucher

Expenses

Record spend against a site or category and post the payment leg — still subject to your approval flow.

get_expensescreate_expense

Trade

Read and raise sales and purchase orders, and pull invoices with their payment status.

get_sales_ordersget_purchase_ordersget_invoicescreate_sales_ordercreate_purchase_order

Inventory, assets & sites

Stock levels and movement per warehouse, fixed asset registers, and the sites everything is booked against.

get_inventoryget_assetsget_sites

People

Headcount, departments, salary structure and payroll runs — only if your ERES role can already see them.

get_employeesget_payrollget_contacts

WhatsApp

Send an invoice, statement or reminder to a party and have the delivery logged in ERES history.

send_whatsapp_message

Data import

Hand the agent a spreadsheet and let it validate, map and load your opening data into ERES.

import_chart_of_accountsimport_bank_accountsimport_data
Security

Giving an agent access should not mean giving up control

The connector was built permissions-first. Everything below is enforced on the server, on every request — not in the prompt.

OAuth 2.1 with PKCE

Authorization-code flow with proof key exchange. Access tokens expire in one hour; refresh tokens rotate on every use and expire in thirty days.

Your password stays yours

The AI client never sees your credentials. It holds an encrypted, revocable token — disconnect from either side and it dies immediately.

Never more than you can do

Every single request is revalidated against your user, company, scopes, assigned modules and write permissions. The agent inherits your role exactly.

Pinned to one organization

The connection is permanently scoped to the organization you authorized. Nothing keyed without a company id, and no crafted id can reach another tenant's data.

Approvals still apply

Anything that posts to the ledger enters your normal approval flow. An agent can prepare work; it cannot skip your controls.

The ERP stays the source of truth

Nothing is cached that was not read from ERES first, every entry has a TTL, and any successful write invalidates what it touched.

Under the hood

Fast because it is engineered, not because it guesses

The MCP layer never touches the database directly. It calls authenticated ERES endpoints, so every rule your accountants rely on is applied exactly once, in one place.

load context

  → resolve missing inputs from it

  → check cache

    hit? return it

    miss? call the ERP → store → update context

  → invalidate what the write touched

One pipeline, every tool

Load context → resolve missing inputs → check cache → call the ERP → store → invalidate what the write touched. No tool implements its own caching, so no tool can get it wrong.

Tuned time-to-live

Reference data caches for an hour, directories for ten minutes, operational registers for two, and anything derived from the ledger for thirty seconds. Writes are never cached.

Conversation memory

The subject of the conversation is carried between calls, so “find supplier Sodexo” → “show the ledger” → “create a payment voucher” names the supplier once. Your explicit argument always wins.

Skills, not guesswork

Fifteen bundled skills teach the agent how ERES actually works — voucher types, the account hierarchy, tax treatment — so it asks the right question instead of inventing an entry.

Point your AI at your actual books

ERES is the first ERP in Pakistan to ship a production MCP server. Bring the agent you already use, and start asking.

OAuth 2.1 + PKCEScoped to your roleRevoke any timeWorks with any MCP client