Finsi MCP

Ask your AI about your store. Get answers from your real numbers.

Connect Claude and other AI agents to your Finsi data over MCP. Read-only, limited to your brand, and every call is logged.

Talk to us
ClaudeChatGPTCodexGrokand any MCP client
The problem

Why AI answers about your store are often wrong

Too many dashboards

Revenue is in Shopify, ads in Meta and Google, email in Klaviyo. Getting one answer means opening all of them.

Hours spent exporting

Pasting CSVs into a chat so an AI can see your numbers is slow, and the numbers are stale by the time you ask.

AI without your context

A general-purpose assistant does not know your metric definitions or your brand, so its answers are guesses.

Finsi is used by commerce brands like these

Harney & Sons
1906 Shop
Caffeine & Legends
History by Mail
PupBox
Vinyl Moon
Introducing Finsi MCP

Your AI reads from Finsi, not from a pasted export

MCP, the Model Context Protocol, is the standard way for an AI assistant to call tools. Finsi publishes its tools over MCP, on top of one data foundation.

Your AI

ClaudeChatGPTCodexGrok
Any MCP client

Finsi MCP

API key, one brand
Read-only
Every call logged

Semantic layer and metrics

One definition per number, calculated once
Data warehouse

Your sources

ShopifyKlaviyoMeta AdsGoogle AdsSubscriptions
  1. 1

    Connect your sources

    Shopify, Klaviyo, Meta, Google Ads and subscription platforms connect to Finsi once.

  2. 2

    One data warehouse

    Finsi brings them into one warehouse, so your AI is not reading five tools with five definitions.

  3. 3

    A semantic layer

    It defines what each number means, so “revenue” or “new customer” is the same for every question.

  4. 4

    Metrics calculated once

    Metrics are calculated with the same definitions across sources, and a catalog lists the ones you can run by name.

  5. 5

    MCP access, under your control

    Your AI reaches all of it through one MCP endpoint: read-only, one brand per key, every call logged.

Why it matters

The same question, with and without Finsi MCP

Without context
With Finsi MCP
Each tool defines revenue and customers its own way, so the answers disagree.
One semantic layer, so every question uses the same definitions.
You export CSVs and paste them into a chat.
Your AI calls Finsi’s tools and reads current data.
The assistant guesses at your brand and your numbers.
It reads your brand profile and your calculated metrics.
Nobody can tell what the AI looked at.
Every tool call is logged with the key, tool and brand.
Finsi in your AI tool

Connect in a few steps

In Finsi, create an agent connection and copy your API key. Then add Finsi to your client.

Claude Desktop

Add Finsi as an MCP server in your Claude Desktop configuration, with your key in the Authorization header.

{
  "mcpServers": {
    "finsi": {
      "transport": "streamable-http",
      "url": "https://api.finsi.ai/mcp",
      "headers": {
        "Authorization": "Bearer finsi_<your_key>"
      }
    }
  }
}

Claude Code

Add Finsi from the command line, then ask questions in your terminal.

claude mcp add --transport http finsi https://api.finsi.ai/mcp \
  --header "Authorization: Bearer finsi_<your_key>"

ChatGPT and other MCP clients

Any client that supports MCP over streamable HTTP with a custom header can connect with the same URL and key. Check your client’s documentation for where the setting lives.

URL:    https://api.finsi.ai/mcp
Header: Authorization: Bearer finsi_<your_key>
See it work

What happens when you ask a question

1

You ask

“Why did our blended acquisition cost change this week?”

2

Your AI picks tools

It calls get_ads_performance for spend and acquisition cost by campaign, and get_revenue_summary for new and returning customers.

3

Finsi returns calculated metrics

The numbers come from the semantic layer, with the same definitions Finsi uses in its own reports.

4

Your AI explains

It answers in plain language and can pull list_ads_recommendations to suggest what to review next. Nothing is changed.

What your AI can read

21 read-only tools in four areas

Brand context

Companies you can access, connected data sources and their status, brand voice and key messaging.

list_companiesget_companylist_brand_data_sourcesget_brand_voice

Analytics

Revenue with new and returning split, orders over time, ads performance (spend, revenue, ROAS, CAC), email performance, top products, customer cohorts and lifetime value.

get_revenue_summaryget_ads_performanceget_email_performanceget_top_productsget_customer_cohorts

Metrics catalog

A list of vetted metric templates, and a tool that runs one by name. Raw SQL is never accepted.

list_metrics_catalogrun_catalog_metric

Reports and recommendations

AI-generated business reports, recommendations across ads, email flows and conversion, and significant events such as anomalies.

list_business_reportslist_business_recommendationslist_ads_recommendationslist_flow_recommendations
Try it

A prompt library to start with

Each prompt maps to tools Finsi exposes today.

Executive summary

“Summarize last week’s revenue, new versus returning customers and ad performance, and point out the significant events.”

get_revenue_summary, get_ads_performance, list_business_events
Media buying

“Which campaigns have the highest ROAS and the lowest CAC over the last 30 days, and what do the open recommendations say?”

get_ads_performance, list_ads_recommendations
Email

“Which email campaigns drove the most attributed revenue recently, and which flows need work?”

get_email_performance, list_flow_recommendations
Retention

“Show cohort retention and lifetime value for the last six months and tell me which cohort stands out.”

get_customer_cohorts
Products

“What are my top products by revenue this quarter, and how did orders trend over the same period?”

get_top_products, get_orders_timeseries
Conversion

“List the open conversion recommendations and what we have already learned from past tests.”

list_cro_recommendations, list_cro_learnings
Copywriting in your voice

“Using my brand voice, draft three subject lines for next week’s campaign.”

get_brand_voice
Control

Built so you stay in control

Read-only

The MCP server only reads. It cannot change ads, flows, listings, prices or anything else. Changes to your money and customers still go through approval in Finsi.

Your key, your brand

You connect with an API key. A key for one brand sees only that brand.

Every call is logged

Each tool call is recorded with the key, the tool, the brand and the outcome. Sensitive values are removed from the log.

No raw SQL

Agents can run only the vetted metric templates in the catalog, not arbitrary queries. Date ranges are capped at 366 days.

Frequently asked questions

1. What is Finsi MCP?

A read-only Model Context Protocol server. It lets Claude and other MCP clients read your Finsi data, such as revenue, orders, ads, email, cohorts, reports and recommendations, so their answers use your real numbers.

2. Which AI tools can connect?

Any client that supports MCP over streamable HTTP with a custom header, including Claude Desktop and Claude Code. Other clients that meet the same requirements can connect too.

3. Can the AI change my ads or campaigns through MCP?

No. The MCP server is read-only. Changes to ads, flows and other live settings go through Finsi with a person’s approval and a record.

4. Is my data safe?

A key is limited to the brand it was created for, access is read-only, agents cannot run raw SQL, and every call is logged. Read our Security and Privacy pages for more.

5. How do I get a key?

Create a connection in Finsi, in the app’s agent connections, and copy the key. If you cannot find it, contact us and we will help.

6. Do I need to use Finsi’s own agents?

No. You can use Claude or another agent you already have, use Finsi’s agents, or both. They follow the same rules.

7. What is MCP?

The Model Context Protocol is an open standard that lets an AI assistant call tools and read data from other systems. Finsi publishes its tools over MCP so your assistant can read your store data.

8. What data can the AI see?

Only what Finsi’s 21 read-only tools return for your brand: brand context, analytics, the metrics catalog, reports and recommendations. It cannot run arbitrary queries.

9. Can I connect more than one brand?

A key for one brand sees only that brand. If you manage several brands, talk to us and we will set it up.

10. Do I need a data team to set it up?

No. You connect your sources in Finsi and add one server to your AI client. If you want help, talk to us.

Put your own AI on your own numbers

Tell us which AI tools you use and we will show you how to connect them.

Talk to us