PrismHR MCP server

prismhr-mcp

The open-source Model Context Protocol (MCP) server for PrismHR. Connect Claude, Cursor, or any MCP-compatible AI agent directly to your PrismHR PEO platform. Automate payroll, benefits enrollment, compliance reporting, AR/billing, carrier EDI files, and Microsoft 365 actions — with verified-schema tools, scope-gated consent, and zero custom integration code.

Built for PEOs, brokers, and payroll operators who want AI agents that actually work against PrismHR — not another brittle script farm.

Maintained by Simploy. MIT-licensed, PyPI-distributed, plugin-friendly. The fundamental layer for PrismHR × agentic AI.

mcp-name: io.github.nikulk2992-jpg/prismhr-mcp

Keywords: PrismHR API, PrismHR integration, MCP server, Model Context Protocol, PEO automation, payroll automation, AI for HR, Claude for PrismHR, agentic AI, benefits enrollment automation, 834 EDI, 401(k) file automation, HRIS AI agent, PEO software integration.

Why this exists

Every PEO running PrismHR ends up with the same Frankenstein stack: Python scripts, Postman collections, Playwright automations, one-off Node apps. Each one re-implements login, session keepalive, retry logic, pagination, and PrismHR's quirks (camelCase schemas, 500 "No data found" gotchas, batch-of-20 caps, silent 401s).

prismhr-mcp centralizes all of that once, as a production-grade MCP server. The AI agent orchestrates; the server owns auth, caching, retries, normalization, and PEO domain logic. Any PEO drops it in and gets a productive AI experience against their own PrismHR tenant — no glue code, no guesswork, no hallucinated endpoints.

Who this is for:

directly into their ops stack

• PEOs running PrismHR who want to wire Claude / Cursor / ChatGPT Desktop

• Benefit brokers and carriers building enrollment automations
• Payroll teams replacing brittle Postman / Playwright workflows
• Consultants shipping PEO AI pilots on tight timelines

Status

Production-ready core. Live on PyPI and the MCP Registry.

scrypt-encrypted disk credential cache, PrismHR session with proactive keepalive (no mid-workflow 401s), automatic refresh on failure, concurrency cap, retry with jittered backoff, 500→empty quirk handling, pagination, batching.

• Auth + session + HTTP client: done. 1Password CLI integration,

no guessed endpoints, no invented fields. 102 response shapes verified and rising.

• Verified-schema gate: every tool grounded in a live UAT probe —

services. meta_call lets the agent invoke any verified method safely.

• 447-method catalog: full PrismHR REST surface indexed across 18

consent store with prerequisite expansion and cascade revoke. Default posture = deny all. Tools enforce scope at call time.

• Connect-time consent system: 15-scope manifest, per-(peo, env) JSON

point at prod PrismHR. Prevents accidental first-run blast radius.

• Production safety gate: PRISMHRMCPALLOW_PROD=true required to

• MCP Registry listed: discoverable by every MCP-aware client.
• Test suite: passing via pytest + respx.

See .planning/architecture.md for the full roadmap and .planning/assistants-roadmap.md for the paid tier details.

What this unlocks for PEOs

Once wired into your PrismHR tenant, an AI agent (Claude, ChatGPT, Cursor) can run any of these workflows against live data — read-only by default, write actions opt-in and scoped.

Daily ops

Shortened. The full README is on GitHub.