How to analyze finance data in PDF files automatically
Export tables from a finance PDF, then run Data Insight, Budget Sentinel and Cashflow Radar locally in Nexus Runner: fraud tests, budget variance, cashflow.
Skills in this bundle
AI Cost Pulse
Compare costs and get model recommendations across 29 AI APIs covering text, image, video, and audio tasks.
Budget Sentinel
Compare budget vs actual spending -- calculates variance, flags over-budget items, supports department grouping and threshold alerts
Buzz Meter
7-day news sentiment scoring for a symbol, exported as a Markdown report (hybrid Finnhub). (Action Co-pilot: generates next-best-actions (actions.json), follow-up tasks and calendar reminders on top of the analysis — what to act on today.)
Cashflow Radar
Forecast cashflow horizons (30/60/90 days), category breakdown, confidence intervals.
Intel Scout
Chinese company intel: registration, funding history, news, related parties, change records. Defaults to free public data (akshare, limited coverage); configure Qichacha/Tianyancha keys for richer data. Use it for due-diligence pre-screening and supplier background checks.
Data Insight
28-mode analysis for Excel/CSV/SQLite: schema inspection, SQL query, statistical summaries, fraud detection (Benford, counterparty, network topology), large-amount test, Journal Entry Testing (JET), business-finance reconciliation (AR/AP aging, inventory turnover), and BI export.
Export the finance tables out of your PDF into Excel or CSV, then run Data Insight for schema checks, statistical summaries and fraud detection, Budget Sentinel for budget-versus-actual variance, and Cashflow Radar for a 30/60/90-day outlook. All three run inside Nexus Runner 1.0.10 on your own machine, reading local file paths and returning exception lists, variance tables and cashflow forecasts.
- Turn the PDF tables into Excel or CSV first; Data Insight then covers 28 analysis modes including schema inspection, SQL query, statistical summaries, fraud detection and Journal Entry Testing.
- Budget Sentinel compares budget against actual spending, calculates variance and share, flags over-budget items, and supports department grouping with threshold alerts.
- Cashflow Radar returns 30/60/90-day cashflow horizons, a category breakdown and confidence intervals.
- For diligence and research work, Intel Scout pulls Chinese company registration, funding history and change records, and Buzz Meter scores 7-day news sentiment for a symbol.
- Skills run on your own PC and data stays offline. Nexus Runner 1.0.10 supports Windows 10/11 x64 only; the free tier includes 200 Nex per month, reset at the start of each month.
What PDF finance analysis looks like on MCP Nexus
MCP Nexus splits finance analysis into skills with explicit inputs and outputs. There are only two kinds: a skill is one schema-bounded operation, an agent handles multi-step orchestration. In this PDF workflow, Data Insight handles inspection and testing, Budget Sentinel owns the budget view, Cashflow Radar owns the cash view, Intel Scout and Buzz Meter fill in counterparty and market context, and AI Cost Pulse prices any model calls the workflow itself makes. Every call happens inside Nexus Runner 1.0.10 on your machine, no file upload, billed in Nex at 3 to 10 Nex per skill run.
From one PDF to an analysis you can archive
How the three approaches differ
| MCP Nexus | Manual work | Generic cloud AI | |
|---|---|---|---|
| Where the data goes | Skills execute inside Nexus Runner 1.0.10 on your PC, reading local paths, so finance tables stay offline | Files move between mailboxes, shared drives and personal laptops with little version or access control | Raw finance tables must be uploaded to a third-party server before anything can be analyzed |
| Fraud and journal testing | Data Insight ships Benford, counterparty consistency, network topology, large-amount testing and Journal Entry Testing | Pivot tables and manual sampling, with the test rules rebuilt each cycle | Mostly narrative summaries, without a test definition you can recompute |
| Repeatability of results | Same input, same output structure on every run, so periods can be filed and compared | Definitions shift with whoever does the work and break at handover | Asking the same question twice can return differently worded or different conclusions |
| Budget to cash handoff | Budget Sentinel and Cashflow Radar take the same table, so variance work feeds straight into the 30/60/90-day view | Budget and cashflow files are maintained separately and reconciled by hand | Output shape varies, which makes it hard to chain into a repeatable process |
| How it is billed | Billed in Nex at 3 to 10 Nex per skill; free tier 200 Nex monthly reset, Pro ¥49/month with 3000 Nex, Enterprise ¥299/month with 50000 Nex | Billed as person-days, rising with overtime in closing season | Subscription or token pricing with no link to how many analyses you actually produce |
Real use cases
FAQ
Can these skills analyze the PDF tables directly?
Export them to Excel or CSV first. Data Insight reads Excel, CSV and SQLite, and once the tables are saved it can run all 28 modes, including schema inspection, SQL query and fraud detection.
Are my finance files uploaded anywhere?
No. Skills run inside Nexus Runner 1.0.10 on your own machine and read local file paths, so data stays offline. Runner supports Windows 10/11 x64 only.
How much Nex does a full PDF finance analysis cost?
Between 3 and 10 Nex per skill run. Buzz Meter is 3 Nex, Cashflow Radar and Intel Scout are 5 Nex each, Budget Sentinel and Data Insight are 10 Nex each. The free tier includes 200 Nex per month with a reset at the start of each month; Pro is ¥49/month with 3000 Nex and Enterprise ¥299/month with 50000 Nex.
Do I need paid data to look up a Chinese company?
Not by default. Intel Scout uses free public akshare data out of the box, with limited coverage. Adding Qichacha or Tianyancha keys gives richer results.
How do I estimate the model cost of the workflow itself?
Run AI Cost Pulse first. It covers API call costs across 29 mainstream models for text, image, video and audio tasks, and turns token counts or usage into a cost comparison table with a recommended best-value option.