Tumiki MCP
JA
← Back to home

Use cases

What MCP actually moves,
with the published numbers

Work that measurably improved with MCP servers and AI agents, taken from primary sources. Every figure below is a published value from the linked source.

What it looks like in use

You ask; Tumiki checks the permission, calls the MCP tool and returns the result. Switch the AI client and the use cases switch with it.

The same MCP servers, from any AI client

Find this quarter's deals more than 14 days past their close date and share them with the owners

Claude
Tools called over MCP (permission checked by Tumiki)
  1. list_opportunitiesAllowedFilter this quarter's deals by overdue close date
  2. post_messageAllowedPost to #sales, mentioning each owner
Done

8 deals are more than 14 days past their close date, worth $820K in total. The oldest is 31 days overdue. I posted the list and mentioned each owner.

8 overdue deals$820K at risk31 days oldest slip
Reply to Claude...

Screens and figures here are an illustration of the flow. They are sample data, not the published results in the cases below.

How to read this page

MCP use confirmed4
The source explicitly describes MCP being used
Reproducible with MCP9
The source does not mention MCP, but the same workflow can be built with MCP servers

Published results come from a combination of the AI model, redesigned workflows, data quality, training and governance — none of them isolate MCP's own contribution. For your own rollout, measure the before-state per workflow and confirm the real reduction in a PoC.

13

Claude connects to Salesforce, Gong and internal data over MCP, so account details, deal history and call content arrive together.

Published results
Deal preparation time
75% reduction
Prep per opportunity
3–4 hours → 45 min
System checks that took 30 minutes
Real time
Claude adoption after deploying MCP servers
7x
MCP servers customers built in 8 months
800+
MCP servers to build it
  • Salesforce MCP
  • Gong MCP
  • Customer data MCP
  • Proposal & knowledge MCP
KPIs to track
  • Deal preparation time
  • Time spent looking up account data
  • Opportunities per rep
  • MCP adoption rate
Source: Anthropic

An MCP connector lets Claude operate the product directly — not only searching, but creating sheets, updating rows and flagging risks from a conversation.

Published results
MCP tool calls in the first 30 days
1.76 million
Organisations using it in 30 days
1,400+
Usage that goes beyond querying data
48%
IT help desk tickets handled automatically
30%
Operational cost
60% reduction
MCP servers to build it
  • Smartsheet MCP Connector
  • IT help desk MCP
  • Project management MCP
  • Auth & permissions MCP
KPIs to track
  • MCP tool calls
  • Share of calls that write data
  • Help desk automation rate
  • Open ticket backlog
Source: Anthropic

Zapier MCP connects Claude to Slack, their CRM, web search and their own codebase, and each team builds its own agents on top.

Published results
AI adoption across all employees
89%
AI agents deployed internally
800+
Tasks processed through Anthropic
10x YoY
MCP servers to build it
  • Zapier MCP
  • Slack MCP
  • CRM MCP
  • Web search MCP
  • Internal codebase MCP
KPIs to track
  • AI adoption rate
  • Number of internal agents
  • Automated task volume
  • Time to complete research
Source: Anthropic

Claude Code runs across development, review, onboarding and incident work. One engineer built the Money Forward Cloud MCP Server in three months.

Published results
Saved per engineer (internal survey)
7 hours/week
API endpoint implementation (70%)
2 days → 5 hours
Code acceptance rate
Over 90%
New developer onboarding
1 week → 1 day
Engineers adopted it (70%+ daily)
Over 80%
MCP servers to build it
  • Money Forward Cloud MCP
  • GitHub / GitLab MCP
  • CI/CD MCP
  • Kubernetes MCP
  • Log search MCP
KPIs to track
  • Hours saved per engineer
  • Implementation lead time
  • Code acceptance rate
  • Onboarding days
Source: Anthropic

Codex automated an incident analysis on a critical system that previously took five experienced engineers three days of digging through logs, code and configuration.

Published results
Critical incident analysis
5 people × 3 days → 30 min
Employees on ChatGPT Enterprise / Codex
~9,000
Reported productivity gains
Over 95%
MCP servers to build it
  • GitHub / GitLab MCP
  • Datadog / CloudWatch MCP
  • Kubernetes MCP
  • Jira / ServiceNow MCP
  • Incident history MCP
KPIs to track
  • Incident analysis time
  • Time to root cause
  • MTTR
  • Engineers pulled into the investigation
Source: OpenAI

Complex Japanese business and financial documents are checked against a checklist and a database of past review examples.

Published results
Review time for complex documents
50% reduction
Productivity gain in testing
Up to 85%
Productivity gain in development
40%
MCP servers to build it
  • SharePoint / Google Drive MCP
  • Document management MCP
  • Checklist & policy MCP
  • Past review examples MCP
KPIs to track
  • Review time per document
  • Missed-issue rate
  • Human correction rate
  • Documents waiting for review
Source: Anthropic

ChatGPT Enterprise and the API went out company-wide for research, reporting, market-data summarisation and product development.

Published results
Release cycle for AI-ready products
3–6 months → ~2 weeks
Time to roll out to thousands of employees
Weeks
MCP servers to build it
  • Financial data MCP
  • Market data MCP
  • Internal research MCP
  • Product management MCP
  • Compliance MCP
KPIs to track
  • Release cycle length
  • Market research time
  • Report writing time
  • Share of answers with cited sources
Source: OpenAI

An M&A agent reads deal, company and financial data. More than twelve separate workflows were consolidated into one general-purpose agent.

Published results
Internal eval accuracy
25% → 85%
Concurrent deals per banker
5–8
Building a presentation
30–40 hours → ~1 hour
Closed in the firm's first year
8 deals / $91M
MCP servers to build it
  • CRM MCP
  • Financial data MCP
  • Company data MCP
  • Deal management MCP
  • Contract & document MCP
KPIs to track
  • Answer accuracy
  • Human correction rate
  • Concurrent deals per person
  • Time to build materials
Source: Anthropic

A sales assistant in Slack reads account history, call notes, Salesforce activity and release updates to produce meeting briefs and product answers.

Published results
Sales rep productivity
20% lift
Time returned to customer work
~1 day/week
Answers to questions that used to linger
Minutes
MCP servers to build it
  • Salesforce MCP
  • Slack MCP
  • Calendar MCP
  • Call recording MCP
  • Product knowledge MCP
KPIs to track
  • Meeting prep time
  • Time spent with customers
  • Time to answer a question
  • CRM update rate
Source: OpenAI

An agent triages thousands of inbound emails a day, routes them, drafts responses and sets handling priority.

Published results
Email response time
60% faster
Time saved with Copilot
19,000 hours/month
Copilot actions taken
340,000/month
MCP servers to build it
  • Outlook / Gmail MCP
  • Zendesk / ServiceNow MCP
  • FAQ & knowledge MCP
  • Customer data MCP
  • Escalation MCP
KPIs to track
  • First response time
  • Average handling time
  • Auto-classification rate
  • Draft acceptance rate
Source: Microsoft

Document classification, information extraction, translation, research drafts and financial-crime investigation all run with AI support.

Published results
Tasks automated in targeted processes
Up to 70%
Reduction in manual processing
Up to 50%
Internal use cases in the first year
15+
From idea to initial evaluation
1–2 weeks
MCP servers to build it
  • Customer data MCP
  • Case management MCP
  • Document management MCP
  • Core banking MCP
  • KYC MCP
  • Fraud detection MCP
KPIs to track
  • Automation rate
  • Manual handling time
  • Handling time per case
  • False positive rate
Source: Anthropic

The legal team uses generative AI to summarise contracts, check terms, surface the points that matter and support decisions.

Published results
Saved per person in legal
4 hours/week
Drafting a new contract
1 hour faster
From trial to company-wide rollout
300 → 68,000
MCP servers to build it
  • SharePoint / OneDrive MCP
  • Contract management MCP
  • Outlook MCP
  • Policy & clause library MCP
  • E-signature MCP
KPIs to track
  • Contract review time
  • Hours saved per person
  • Days to contract signature
  • Missed clause rate
Source: Microsoft

AI covers catching up on meetings, writing minutes, gathering information and audit analysis.

Published results
Saved on meetings and minutes
9,000+ hours
Internal audit team efficiency
30% gain
MCP servers to build it
  • Teams MCP
  • Outlook MCP
  • SharePoint MCP
  • BI & spreadsheet MCP
  • Audit document MCP
KPIs to track
  • Time to write minutes
  • Time to log follow-up tasks
  • Internal audit effort
  • Missed task rate
Source: Microsoft

Where the gains were largest

Each row is a published figure from the cases above. The company is named alongside it.

WorkflowPublished result
Incident analysis & logs5 engineers × 3 days → 30 minNTT DATA Group
Deal prep & account research3–4 hours → 45 min (75%)Workato
Implementation & API work2 days → 5 hours (70%)Money Forward
Routine back officeUp to 70% of tasks automatedN26
Inbound supportEmail response 60% fasterCapita
Document reviewReview time cut 50%Nomura Research Institute
Contract review4 hours/week per lawyerVodafone
Developer onboarding1 week → 1 dayMoney Forward
Help desk30% of tickets handled automaticallySmartsheet
Product release3–6 months → ~2 weeksLondon Stock Exchange Group

Where MCP is worth adding first

The cases above share one of these three shapes.

The work spans several systems

Once AI can reach each system, the searching, copying and stitching disappears with it.

  • CRM, email, chat and drive to prepare for a meeting
  • Repo, logs and issue tracker to investigate an incident
  • Contracts, clause library and past cases for legal review

Each item follows a fixed procedure

When the steps are clear, MCP can handle the updates and filing too, not just the lookups.

  • Triaging inbound email
  • Matching invoices
  • Logging tasks after a meeting
  • First-response checks on an incident

Finding internal data takes too long

Company-specific context that AI cannot answer on its own, fetched within the caller's permissions.

  • Account history
  • Past incidents
  • Internal policy
  • Product specs
  • Project history

How to measure it

Compare the same conditions before and after, per workflow.

MetricFormula
Time reduction(before − after) ÷ before
Automation ratecases completed by AI alone ÷ all cases
Draft acceptanceoutputs a human kept ÷ all AI outputs
Error reduction(errors before − errors after) ÷ errors before
Adoptionmonthly MCP users ÷ eligible users
Write ratecalls that created or updated data ÷ all MCP calls
ROI(annual benefit − annual cost) ÷ annual cost

Worked example

This is a calculation from assumed inputs, not a published result.

  1. 50 reps, 10 opportunities each per month, 2 hours saved per opportunity:
  2. Monthly hours saved = 2 × 10 × 50 = 1,000 hours
  3. At a $35 blended hourly rate: 1,000 × $35 = $35,000 per month
  4. $420,000 a year — then subtract AI licences, MCP build, maintenance and training to get ROI

Do not stop at search

The larger gains come from handing over the whole chain: fetch the data, decide, update the system, notify the people involved. In the Smartsheet case, 48% of MCP usage was creating and updating rather than querying. A read-only rollout and a rollout that can act are not the same investment.

Try it on your own workflow

Tell us the workflow and we will map the MCP servers and the permission model it needs.

Talk to usTry the live demo