MCP Claude · Cursor · Claude Code · 7 HR modules · audit-logged

Your HRMS, in
any AI tool.

Graciax HRMS runs a Model Context Protocol server. Connect the AI you already use and it can check who is in, apply leave, pull a payslip, run a headcount report or draft an onboarding process, with your permissions, confirmed before every write, and logged every time.

Book a Demo All HRMS Features
One API key per person Keys inherit your role Writes confirm first
Claude · connected to Graciax HRMS · key: Anjali (HR)
YOU
Which contracts expire in the next 60 days, and who is finishing probation this month?
CLAUDE get_contract_expiring_soon get_probation_ending_soon
2 fixed-term contracts end before 31 Oct: Sameer K. (Operations) and Divya R. (Finance). 3 people finish probation this month; two have no review scheduled yet.
YOU
Draft an onboarding process for field sales: IT setup, device issue, buddy assignment, 30-day review.
CLAUDE create_workflow_draft add_workflow_step add_workflow_edge
Draft “Field sales onboarding” v1 created with 4 steps and a trigger on employee created. It is not published. Want me to summarise it so you can decide?
  • 5 tool calls logged against Anjali · scope: HR
  • Publishing needs the workflow.publish scope and your explicit yes
Model Context Protocol

What MCP is, in plain words.

MCP is an open standard for connecting an AI client to a live system. The system publishes a list of tools it can do, such as get_leave_balances or create_reimbursement. The AI reads the list, picks the right tool for your question, calls it, and answers from the result. Graciax HRMS runs an MCP server, so one integration serves every AI client that speaks the standard.

  • Open standard. Not locked to one AI vendor
  • Works with Claude, Cursor, Claude Code and any MCP-aware client
  • The AI calls Graciax tools. It gets back only what your role is allowed to see
  • The same tools power the in-app AI assistant, so behaviour is identical
CClaude
API key · Anjali (HR)Connected
⊞Cursor
API key · Dev team (Employee)Connected
▸Claude Code
API key · Meera (Manager)Connected
7 modules exposed · every call audit-logged
Each key sees a different HRMS. Anjali’s sees everyone. Meera’s sees her team. The Dev team’s keys see their own records.
How one question flows

Ask. Call. Answer. Log.

You ask in your AI client

“How many casual leaves do I have left?” typed into Claude, Cursor or Claude Code, exactly as you would ask a colleague.

The AI calls a Graciax tool

It picks get_leave_balances from the published tool list and calls the Graciax MCP server with your key. Your role decides what comes back.

You get an answer, HR gets a log

The AI replies from live data. The call is recorded with your name, the tool, the time and the outcome. Writes would have asked you first.

Seven modules

Real tool names.
Example prompts that work.

Each module maps to a set of MCP tools. The prompts below are the kind of thing employees, managers and HR actually type. Reads answer at once. Writes confirm first.

Workflow authoring“Draft an onboarding flow for field sales with a 30-day review step.”

Read your published org processes, draft new versions step by step, wire triggers and approvals. Publishing needs its own scope and an explicit yes after the AI summarises the diff. Terminations and payroll finalisation can never be published this way.

list_workflow_processescreate_workflow_draftadd_workflow_stepadd_workflow_edgeadd_workflow_triggerpublish_workflow_version
Read processes
Draft new versions (workflow.draft scope)
Publish after a diff and your yes (workflow.publish scope)
Publish termination or payroll finalisation via MCP
Enterprise plan

Leave & attendance“Apply 2 days casual leave from Monday and show my balance.”

Balances, who is in today, upcoming leaves, the team calendar, punches and anomalies. Managers approve or reject from the same chat.

get_leave_balancescreate_leave_requestget_who_is_in_todaydecide_leave_request

Payroll & reimbursements“Show my August payslip and submit a cab fare claim.”

Payslips and salary structure for yourself, payroll runs and cost reports for HR and admins. Claims are submitted with a confirmation.

get_payslipcreate_reimbursementget_payroll_summaryget_payroll_cost_report

Performance“What are my goals this cycle? Submit my self-rating.”

Cycles, goals, appraisal status and self-ratings. HR can submit on behalf of someone when a cycle is closing.

get_performance_goalsget_performance_cyclessubmit_self_ratingget_appraisal_status

Team & org“Who reports to Meera? Find the leave policy.”

Directory search, org chart, team members, shift schedules and organisation documents such as policies and handbooks.

search_employeesget_org_chartget_team_membersget_org_documents

Documents & compliance“Find my offer letter. What is due this month?”

HR letters, employee documents, the compliance calendar, expiring contracts, documents and work permits, and open alerts.

get_hr_lettersget_compliance_calendarget_document_expiry_alertsget_contract_expiring_soon

Manager reports“Attrition this quarter, headcount by department, open pipeline.”

Headcount, attrition, new joiners, recruitment pipeline, pending tasks and approvals. Numbers come from the live record, not a stale export.

get_headcount_reportget_attrition_reportget_recruitment_pipelineget_pending_approvals
Trust

AI capability without AI risk.

Keys & audit trail

One key per person. One log for everything.

An admin creates API keys in settings. Each key belongs to a person and carries that person’s role, so there is never an anonymous AI acting on your HRMS. Every call, from the in-app assistant or any MCP client, lands in an audit log admins can read at any time.

  • Rotate or revoke any key from settings, instantly
  • Every call records user, tool, timestamp and outcome
  • Writes ask for confirmation before they execute
  • Bank account, Aadhaar and PAN numbers are never returned in full
10:42 · Meera · decide_leave_request · approved #1042
10:31 · Arjun · get_payslip · own record
09:58 · Anjali via Claude · get_headcount_report
09:40 · Dev team key · get_salary_structure · denied, out of scope
3 active keys · last rotation 12 days ago
Denied calls are logged too. A key that asks for more than its role allows gets a refusal, not data, and the attempt is on record.
Role scoping

The AI is exactly as powerful as the person holding the key.

There is no separate AI permission model to maintain. Whatever a person can see and do in the web app, their AI can see and do, and nothing more. Change the person’s role and every key they own changes with it.

  • Admin keys see the whole workspace
  • HR keys see people data, letters, compliance and reports
  • Manager keys see and approve for their own team
  • Employee keys see their own leave, attendance, payslips and goals
ADAdmin
Workspace · keys, scopes, audit logAll
HRHR
People, payroll, letters, compliance, reportsOrg
MGManager
Own team · approvals, who is in, team reportsTeam
EMEmployee
Own leave, attendance, payslips, goals, documentsSelf
Salary stays private. CTC and salary structure are shown only to the employee themselves and to HR or admin roles.
Workflow authoring guardrails

The AI drafts your processes. A human publishes them.

Onboarding, offboarding, approvals, reviews: the AI can read what you run today and draft a better version. Getting a draft live is deliberately a two-step, human decision.

  • Read is the default. Listing and inspecting published processes needs no extra scope.
  • Drafting needs the workflow.draft scope, switched on by the super admin. The AI creates a new version rather than touching the one that is live.
  • Publishing needs the workflow.publish scope and a confirmation. The AI must first call without confirming, show you the diff, get your explicit yes, then call again.
  • Sensitive processes never publish over MCP. Terminations and payroll finalisation always need a person in the web UI.
list_workflow_processes, get_workflow_process
create_workflow_draft, add_workflow_step (workflow.draft)
publish_workflow_version, confirmed twice (workflow.publish)
Publish termination or payroll finalize
Claude · Graciax HRMS · workflow.publish enabled
YOU
Publish the field sales onboarding draft.
CLAUDE publish_workflow_version
Before I publish, here is v1 against nothing live today:
  • Trigger: employee created, department = Field Sales
  • Steps: IT setup → device issue → buddy assignment → 30-day review
  • Approver on step 4: reporting manager
Shall I publish this as v1? Reply yes to confirm.
YOU
Yes.
CLAUDE publish_workflow_version confirmed=true
Published Field sales onboarding v1. New joiners in Field Sales will start on it from now. Logged against Anjali.
Set-up

Connected in four steps.
No IT project.

1

Create a key

An admin opens Settings, creates an API key for a person and, on Enterprise, switches on the workflow scopes they should have.

Settings · API keys
2

Add Graciax to your client

Paste the key into Claude, Cursor or Claude Code as an MCP server. The client loads the tool list for all seven modules.

Claude · Cursor · Claude Code
3

Ask

Balances, who is in, payslips, policies, reports. Reads answer at once with your role applied.

read tools
4

Act, with a yes

Apply leave, approve a request, submit a claim, draft a process. The AI confirms first, then acts, then logs it.

write tools · confirm first

Prefer not to leave the app? The same tools power the in-app AI assistant for employees, managers and HR. No key needed.

Questions

Straight answers about MCP for HR

What is MCP?
Model Context Protocol is an open standard that lets an AI client such as Claude or Cursor call tools in a live business system. Graciax HRMS runs an MCP server, so the AI you already use can read and act on your HR data with your permissions.
Which AI tools work with Graciax HRMS?
Claude, Cursor, Claude Code and any MCP-aware client. An admin creates an API key in settings and you paste it into the client. The key inherits the role of the person who created it.
What can the AI change, and what can it only read?
Reads such as balances, who is in, payslips and reports are instant. Writes such as applying leave, approving a request or submitting a claim are confirmed with you before they run. Publishing a workflow needs a separate scope and an explicit yes after a diff. Terminations and payroll finalisation can never be published through MCP.
Who can see what?
Admins see the workspace, HR sees people data, managers see their team, employees see their own records. Salary is shown only to the employee and to HR or admins. Bank account, Aadhaar and PAN numbers are never returned in full.
Is there an audit trail?
Yes. Every tool call is recorded with the user, the tool, the timestamp and the outcome, whether it came from the in-app assistant or from an MCP client. Admins can review it at any time and revoke or rotate keys from settings.
Which plan includes MCP and workflow authoring?
Workflow authoring over MCP is an Enterprise plan feature. Ask Sales which plan fits your team and they will size a custom quote.
Explore more

What the AI is reading from

AI & MCP is one of 17 modules. Every one is connected to the same employee record.

Reports & Analytics

16+ ready reports including headcount, attrition, payroll register, EPF, ESI and TDS. Branded HTML and CSV export.

Statutory Compliance

EPF, ESI, TDS, PT (15 states), LWF (13 states), Bonus, POSH, Form 16, Form 24Q, F&F with gratuity.

Employee Self-Service

Punch in and out, apply leave, download payslips and file tax declarations. All in-app, with the AI assistant alongside.
Get started

Plug your AI into HR
this week.

Book a demo and watch Claude answer a leave question, pull a headcount report and draft an onboarding process on a real Graciax workspace. Then ask for a custom quote.

Book a Demo Start Free