What Is the Numeric MCP?
Last updated: May 9, 2026
This is Part 1 of the Numeric MCP series. Whether you're evaluating the MCP for your team or just heard about it for the first time, start here.
Pick your path:
Already know what the MCP is and want to connect it? Jump to Part 2: Getting Started →
Ready to install pre-built workflows? Skip to Part 3: Skills & the Toolkit →
Want to see automation in action? Go straight to Part 4: Accruals & Journal Entries →
The Short Version
Your Numeric workspace already holds everything you need to run the close — tasks, reports, GL data, commentary, and history. The Numeric MCP is what lets an AI assistant like Claude reach into that data and actually do something with it.
MCP stands for Model Context Protocol. It's a controlled connection that lets an AI platform, like Claude, read data from Numeric, take actions inside Numeric, and write results back — all while respecting the same permissions you already have in the app.
Three Layers, One System
Think of it as three layers working together:
Numeric is the operating system. It holds ownership (who's assigned to what), rules (how work runs), history (what happened last period), and financial data (balances, transactions, reports). This is your source of truth — that doesn't change.
The Numeric MCP is the interface. It exposes Numeric's live state to AI and writes results back. It's the controlled handshake: Numeric decides what data is available and what actions are allowed, and the MCP enforces that on every request.
Claude is the execution brain. It applies judgment, collaborates across systems, and operates in plain language. You tell it what you need, and it figures out how to get there using the tools the MCP provides.
The key takeaway: Numeric stays the source of truth. Claude does the heavy lifting. The MCP connects them.
What Can the AI See and Do?
The MCP gives Claude access to a specific set of tools, split into two categories.
Read tools — what the AI can see:
Workspace and close period context
Close task management (assignments, statuses, due dates)
Financial reporting and GL data
Flux explanations and transaction lines
Task comments and event history
Write tools — what the AI can do:
Create, edit, assign, and submit tasks
Set due dates and add task comments
Update flux explanations
Categorize tasks and create properties
Build and retrieve reports
Your Claude administrator controls which tools are enabled. Every action is logged in Numeric's activity history, so you always have a complete audit trail.
Security and Governance
A natural first question: Is this safe for my financial data?
SOC 2 covered. The MCP server runs inside the same environment Numeric's SOC 2 Type II program covers. It will be in scope for the 2026 annual audit and pen test.
Permissions are inherited. The MCP enforces your existing Numeric role on every request, server-side. A staff-level user can't do anything through Claude that they couldn't do in the Numeric UI. Authentication is per-user via OAuth.
You control the write surface. Your LLM administrator can restrict the Numeric connector to read-only at the workspace level. Individual users can disable specific tools or require manual approval before any write action. Access is revocable instantly from either side.
Numeric doesn't see your prompts. Numeric does not retain or log the prompts you send to your LLM. It only provides data when the AI is explicitly prompted to call a specific tool.
What Does This Unlock?
The MCP opens up capabilities across three domains:
Insights — close period pulse checks, account anomaly detection, financial metrics, task assignment optimization, and close retrospectives.
Automation — accrual identification, journal entry posting, overdue task reminders, bulk edits, template preparation, and automated communications via tools like Slack and email.
Workflows — consolidated reports, executive summaries, custom reporting dimensions, audit evidence exports, supporting workbooks, aging analysis, and bespoke dashboards.
Some of these are single prompts. Others chain multiple tools across Numeric, your ERP, Slack, and Google Calendar into workflows that replace 10–20 minutes of manual work. We'll get into specifics in Parts 3 and 4.
Where Are You on the Adoption Curve?
AI adoption is a spectrum. Knowing where you sit helps you pick the right next step.
Exploring — You've read about MCP and maybe watched a demo. No one has tried it on the close yet. You're watching what peers are doing. You're in the right place — keep reading this series.
Activating — One person is already exploring tools. Claude access and IT review are in motion. A first workflow is identified, ready to launch. Head to Part 2 and get connected.
Operating — One or two workflows are running steadily. A champion has built the habit with saved prompts and repeatable patterns. It hasn't spread beyond the original use case yet. Jump to Part 3 and expand your toolkit.
Scaling — Multiple teammates are using MCP daily. Custom skills are being built in-house. Cross-tool workflows span Slack, Google Drive, and your ERP. Part 4 shows what end-to-end automation looks like at this stage.