Numeric's MCP Server: How to Connect and 11 Key Use Cases

The Three Levels of Using Numeric’s MCP

Beginner: Take actions directly in Numeric

At the most basic level, Claude becomes a natural language interface to your Numeric workspace. 

The Numeric MCP server exposes 20+ tools organized into three domains.

1. Workspace and Context - AI can orient itself in your Numeric environment - understanding who’s on the team, what periods are open, which entities exist, and what tags and categories are available.
2. Close Task Management - your AI model can participate in the full lifecycle of a close task: it can list tasks, create tasks, assign and reassign preparers and reviewers, set due dates, add comments, etc.
3. Financial Reporting & GL Data - have AI pull structured financial data, build ad-hoc reports, analyze trends, and drill into individual journal entries.

From here, you can do virtually all of the same actions you’d normally perform in Numeric or pair tasks together for new, multi-step capabilities:

• Query your GL in plain English - "What did we spend on AWS last quarter?" Claude builds an ad-hoc income statement, pivots by vendor, filters to the relevant account, and returns the answer in seconds. 
• Get real-time close status - "Which tasks are overdue right now?" Claude can pull the current task list, groups by status, and generate a close status dashboard as needed.
• Edit tasks in bulk, with guardrails - Need to push all due dates back three days? Reassign 15 tasks from someone who's out? Claude shows you exactly what will change before executing. You can preview before committing on every bulk operation.‍
• Build, format, and export reports with ease - Build IS and BS reports on the fly or pull any saved report from Numeric and ask Claude to reformat it — add commentary, restructure the layout, produce a board-ready summary. Then, export.

Intermediate: Use cross-platform automations with Slack, Gmail, etc.

The more teams set up MCPs across their tech stack, the more each individual MCP becomes powerful - now able to interact with other tools in ways that once required shaky API calls or sustained engineering resources.

Connect it to tools like the MCPs for your calendar, Slack, and Gmail to create useful communication automation deploy agents directly against your checklist tasks. By combining the Numeric MCP with other productivity tools, even teams that once closed by 5BD can reach new levels of automation.

Slack + Numeric

• Daily close standup bot: Each morning during close, Claude checks all open tasks for the current period, identifies what's overdue or due today, and posts a formatted summary to your #accounting-close channel.‍
• Targeted DMs for overdue tasks: Claude identifies which tasks are approaching or past due, then messages the assignee via Slack with context: "Hey — the Prepaid Amortization task for March is due tomorrow and still in prep. Need help unblocking?" Task name, status, and reviewer comments are pulled directly from Numeric.
• Variance alerts: Claude runs a monthly comparison against the prior period. If any account moves more than a threshold you set (say, 15%), it posts to a channel such as #finance-alerts.

Google Calendar + Numeric

• Auto-block close windows: At the start of each period, Claude reads task due dates and creates calendar blocks for each assignee. Reviewers get blocks after their preparer's due date. If dates shift in Numeric, the events update.

Gmail + Numeric

• Controller/CFO status emails: Claude drafts a weekly close status email to your CFO or Controller with completion rates, overdue items, and returned tasks, pulled live from Numeric. The draft lands in Gmail. You review and send.

Advanced: Multi-level actions that actually do the work

At this level, Claude completes entire accounting workflows that would otherwise take hours. It's not just moving data around; it's producing deliverables. Turn any report or flux analysis into a slide deck, live dashboard, or executive summary with just one prompt.

This is where the value of a well-structured close process compounds. Every task, every workflow, every piece of documentation you've built in Numeric becomes context that agents can reason over and equip in its actions. The more you've invested in organizing your close, the more your agents can do — and because agents read and write directly to your Numeric checklist, Numeric serves as the audit trail for everything they touch.

Budget vs. actuals analysis

Compares a static budget file against live actuals from Numeric to surface material variances and draft written explanations:

• Budget ingestion: reads a user-uploaded budget file (Excel, CSV, or Google Sheets) and maps line items to Numeric's chart of accounts via list_financial_accounts. Flags any unmapped accounts for user review before proceeding
• Actuals pull: queries Numeric via build_report or get_report_data to pull actuals for the matching period, account structure, and entity. Supports monthly, quarterly, or YTD comparisons
• Variance computation: calculates dollar and percentage variances by account, department, or any pivot dimension available in the report. Separates favorable from unfavorable variances and ranks by materiality
• Materiality filtering: applies a user-defined threshold (e.g., >$10K and >5%) to surface only the variances that matter. Defaults to a reasonable threshold if none is specified
• Driver analysis: for flagged variances, drills into underlying transaction lines via query_transaction_lines to identify specific drivers — new vendors, headcount changes, one-time charges, timing differences
• Narrative generation: drafts written variance explanations per account or account group, ready for controller review. References the specific drivers identified and quantifies their contribution to the total variance
• Flux writeback: optionally writes the variance explanations back to Numeric via update_flux_explanation so they're preserved alongside the report data
• Output: formatted variance report (Excel or structured summary) with account, budget, actuals, variance $, variance %, and narrative explanation per line — plus optional writeback to Numeric

Accruals bot

Automates the end-to-end accruals workflow across Google Drive, your GL, and Numeric — from calculating the accrual to posting the entry to closing the task:

• Historical pull: queries Numeric via build_report or get_report_data to pull the trailing 6-month actuals for the target accounts (e.g., utilities, software, professional services)
• Rolling average calculation: computes a rolling average across the lookback window to generate the current-month accrual estimate. Supports simple average, weighted average, or user-specified logic
• Schedule update: writes the updated accrual amounts into the team's accruals schedule in Google Drive (Sheets or Excel), appending the new period column and preserving prior-month history
• Journal entry posting: formats the accrual as a journal entry and posts it to NetSuite (or your GL) via its MCP — debit expense, credit accrued liability, with memo referencing the calculation method and period
• Task completion: locates the corresponding accruals task in Numeric via list_tasks, adds a comment summarizing the accrual amount, method, and GL entry reference via add_task_comment, then submits the task for review via submit_task
• Variance flag: compares the current-month accrual to prior month. If the delta exceeds a threshold (e.g., 10%), flags it in the task comment for reviewer attention before submitting
• Output: updated Google Sheets schedule, posted GL entry, and completed Numeric task with full audit trail — one prompt, three systems

Board package generation

Assembles a multi-report financial package into a single formatted Excel workbook for board, parent-company, or audit delivery:

• Report discovery: runs list_reports to find all saved report configs, matches user intent to the right set (e.g., “board package” → IS + BS + CF; “just the income statements” → IS only)
• Multi-entity support: pulls report data per entity via get_report_data, creates one sheet per entity per statement type (e.g., “Interstellar IS”, “Saturn IS”, “Consolidated IS”)
• Statement types: handles Income Statement, Balance Sheet, Cash Flow, and custom reports. User can specify which to include or default to all three
• Consolidated tab: optionally generates a consolidated view that sums across entities, with intercompany elimination if flagged by user
• Formatting: applies financial statement formatting per sheet — proper headers, account hierarchy indentation, subtotal underlines, currency symbols, $K/$M scaling, negative values in parentheses
• Cover sheet: generates a summary tab with table of contents, entity list, period covered, and generation timestamp
• Bulk mode: can also operate as a simple bulk export — “export all 51 reports as individual CSVs” or “export these 5 reports into one workbook, one sheet each” without the board formatting
• Output: formatted .xlsx via xlsx skill, or individual files (CSV/TSV) if user prefers raw data

Consolidated flux across entities

Consolidates flux variance across multiple dimensions into a single unified view:

• Across entities: merges flux tasks and commentary from multiple subsidiaries into one consolidated variance narrative. Pulls entity-specific flux tasks via list_tasks filtered by report_id, then get_task_comments per task.
• Across flux reports: combines flux tasks scattered across multiple reports (MoM IS, QoQ IS, MoM BS, QoQ BS) into a single queue per account — so a controller sees one view of all variance work regardless of which report it lives in
• Across accounts (roll-up): takes child-account-level flux explanations and summarizes them into group-level narratives (e.g., 12 individual OpEx account explanations → “OpEx up 8% driven by headcount additions in R&D and a one-time legal settlement in G&A”)
• Across periods (trend): stitches flux commentary on the same account across 3-6 months via get_report_row_trend into a trend narrative (e.g., “Software Expense has increased 3 consecutive months, driven by new vendor onboarding”)
• Outputs as structured summary with group account, consolidated variance $, variance %, and rolled-up commentary — ready for CFO review or board prep

Auditor prep packet

Compiles a SOX-ready audit trail from task activity across checklist, recon, and flux modules:

• Activity log: iterates list_tasks → get_task_events per task to build a structured timeline: timestamp, user, action type (created, assigned, submitted, reviewed, commented, reopened), task/account, old value → new value
• Segregation of duties: validates that preparer ≠ reviewer using event data. Flags any tasks where the same person submitted and approved
• Timeliness evidence: compares actual completion timestamp against due date. Flags late completions with days-overdue count. Summarizes on-time vs. late rates per assignee and per report
• Completeness gaps: identifies tasks missing expected lifecycle events — submitted but never reviewed, still PENDING past due date, assigned but no activity
• Comment trail: pulls get_task_comments alongside events so supporting documentation is bundled with the activity log in one export
• Period filtering: scopes the export to a specific close period so auditors get only the month under review
• Outputs as formatted Excel (.xlsx) with columns: period, report, task/account, action, user, timestamp, due date, days late, SoD flag, comments

Numeric MCP Tool List

As it stands, here are the tools that users can access via Numeric MCP:

What’s Next

The Numeric MCP is live now. Connect at https://api.numeric.io/mcp and start with any of the prompts above, or ask Claude to look around your workspace and go from there.

If you try something that works well, or find a workflow that's missing, we want to hear about it. The best recipes will become first-class templates that ship with the product — and the most-requested capabilities will shape what we build in the server itself.

The close has always been a process that demands both precision and speed. The MCP doesn't change what precision requires. But it certainly can change what speed looks like.