AI Analyst Plugin

Ask a business question in plain English. Get validated findings, publication-quality charts, and a slide deck.

Install

Claude Cowork

1. Open Claude Cowork and start a Cowork session
2. Click the + button in the chat box
3. Click Plugins
4. Click Add Plugin
5. Click Personal
6. Click the + button
7. Click Add marketplace from GitHub
8. Paste this URL: https://github.com/Amar1404/AI_ANALYST
9. Click Sync

That's it — the AI Analyst plugin is now available in your Cowork sessions.

Claude Desktop (via Marketplace)

1. Open the Claude Desktop app
2. Open Settings → Plugins
3. Click Add Marketplace
4. Select Add marketplace from GitHub
5. Paste this URL: https://github.com/Amar1404/AI_ANALYST
6. Click Sync
7. Find AI Analyst Plugin in the marketplace list and click Install
8. Restart Claude Desktop when prompted

The plugin's skills and agents are now available in any Claude Desktop conversation.

Getting Started

First time? Say "help me set up" to connect your data and configure your preferences.

Try it: Just ask a question like "Which channel has the highest churn and why?" or "Show me MRR trends over time"

What You Get

• 37 skills — question framing, metric definition, data exploration, forecasting, experiment design, and more
• 20 agents — specialized analytical agents for hypothesis testing, cohort analysis, root cause investigation, storytelling, chart-making, and deck creation
• Publication-quality charts — styled visualizations with consistent theming
• Slide decks — Marp-powered presentations ready for stakeholder readouts
• Data validation — built-in confidence scoring, source tieout, and logical validation

Integrations

The plugin ships with three MCP servers that connect to your data and business context. Configure them in Settings → Plugins → AI Analyst Plugin after install.

Athena (Data Warehouse)

Query AWS Athena tables directly from any Claude conversation.

Required settings:

Prerequisites:

• AWS CLI installed and aws configure --profile <name> completed
• IAM permissions: athena:*, glue:GetTable*, glue:GetDatabase*, s3:GetObject + s3:PutObject on the staging bucket

How to configure (step-by-step):

1. Open Claude Desktop → click the gear icon (top-right) → Settings
2. In the left sidebar, click Plugins
3. Find AI Analyst Plugin in the installed list → click the ⚙ Configure button next to it
4. Scroll to the Athena section. Fill in each field:
   • AWS Profile → paste the profile name from your ~/.aws/credentials file (e.g., default or org-prod)
   • Athena Database → the default database queries should target (e.g., default)
   • Athena S3 Staging → the full s3://... path where Athena writes query results. Must end with /
   • Athena Region → AWS region of your Athena workgroup (e.g., ap-south-1)
   • Athena Workgroup → workgroup name (use primary if unsure)
5. Click Save
6. Restart Claude Desktop (Cmd+Q then reopen) — MCP servers only pick up env-var changes on restart

Start it: Ask "list athena tables" or "sample the orders table" — the athena MCP tools (list_athena_tables, describe_athena_table, query_athena, sample_table) become available automatically once the plugin loads.

Superset (BI Dashboards)

Connect to a Superset instance to read dashboards, charts, and run SQL via the Superset API.

Required settings:

How to configure (step-by-step):

1. (One-time) Create a Superset API key:
   • Log into Superset in your browser
   • Click your profile icon (top-right) → Settings → API Keys (or Security → API Keys on older versions)
   • Click + Create, give it a name (e.g., ai-analyst-plugin), and copy the generated key (starts with sst_...). You will not be able to see this key again.
2. (One-time) Find your Database ID:
   • In Superset, go to Data → Databases
   • Find the database row you want as default → note the numeric ID in the URL when you click into it (e.g., /databaseview/edit/1 → ID is 1)
3. Open Claude Desktop → Settings → Plugins → AI Analyst Plugin → ⚙ Configure
4. Scroll to the Superset section. Fill in:
   • Superset URL → full base URL, no trailing slash (e.g., https://superset.example.com)
   • Superset API Key → paste the sst_... key from step 1 (recommended)
   • Superset Username / Password → leave empty if using API key; otherwise enter DB-auth credentials
   • Superset Auth Provider → db for API key or DB auth; ldap or oauth if your instance uses those
   • Superset Database ID → the numeric ID from step 2
   • Superset Schema → default schema name (e.g., default)
5. Click Save, then restart Claude Desktop

Start it: Leave Superset URL empty to skip this integration entirely. Otherwise, the superset MCP tools activate on plugin load — ask "list superset dashboards" or "run this query against superset" to use them.

OpenMetadata (Live Schema & Descriptions) — optional

When configured, OpenMetadata becomes the preferred live source for table/column schema and descriptions: schema and description changes made in OpenMetadata show up immediately, with no manual edits to the knowledge repo. The knowledge repo's schema.md remains the offline fallback. Quirks, metric definitions, golden queries, and PII/partition guardrails always come from the Knowledge Repo — OpenMetadata does not replace those.

This connects to OpenMetadata's own built-in MCP server (v1.12+), so there is nothing to run locally beyond a small proxy.

Prerequisite: Node / npx must be installed. The connection uses the mcp-remote proxy, which is fetched automatically on first run via npx -y. (The other integrations are pure Python; this one needs Node.)

Required settings:

How to configure (step-by-step):

1. (One-time) Generate a Personal Access Token:
   • Visit <YOUR-OpenMetadata-SERVER>/users/<YOUR-USERNAME>/access-token in your browser
   • Click Generate New Token and copy it
2. Open Claude Desktop → Settings → Plugins → AI Analyst Plugin → ⚙ Configure
3. Scroll to the OpenMetadata section. Fill in:
   • OpenMetadata URL → full base URL, no trailing slash (e.g., https://openmetadata.example.com)
   • OpenMetadata Personal Access Token → paste the token from step 1
4. Click Save, then restart Claude Desktop

Start it: Leave OpenMetadata URL empty to skip this integration entirely — the Knowledge Repo schema.md stays the primary schema source. Otherwise, the semantic_search (meaning-based table discovery), search_metadata (keyword lookup), and get_entity_details tools activate on plugin load and become the preferred schema source automatically.

Knowledge Repo (Business Context)

A GitHub-backed (or local) knowledge base of dataset schemas, metric definitions, business glossary, and SQL patterns. The plugin reads this on every question to ground its answers in your org's reality.

Required settings — choose ONE source:

Expected repo layout:

knowledge-repo/
├── datasets/
│   └── <dataset-id>/
│       ├── schema.yaml
│       ├── metrics.yaml
│       └── glossary.md
└── organizations/
    └── <org>/...

How to configure (step-by-step):

Pick one source — GitHub-backed (recommended for teams) or local (for testing).

Option A — GitHub-backed knowledge repo:

1. (One-time) For private repos only: create a GitHub Personal Access Token:
   • Go to https://github.com/settings/tokens → Generate new token (classic)
   • Scope: check repo
   • Copy the ghp_... token. Public repos can skip this step.
2. Open Claude Desktop → Settings → Plugins → AI Analyst Plugin → ⚙ Configure
3. Scroll to the Knowledge section. Fill in:
   • Knowledge Repo URL → full HTTPS or SSH URL (e.g., https://github.com/test/knowledge-repo)
   • Knowledge Repo Branch → branch name (defaults to main)
   • Knowledge GitHub Token → paste the ghp_... token (private repos only; leave empty for public)
   • Knowledge Datasets → comma-separated dataset folder names (e.g., org_name). Leave empty to auto-discover all datasets/*/ folders.
   • Knowledge Local Path → leave empty
4. Click Save, then restart Claude Desktop

Option B — Local knowledge directory (for testing):

1. Clone or create your knowledge repo locally (e.g., git clone <url> ~/knowledge-repo)
2. Open Settings → Plugins → AI Analyst Plugin → ⚙ Configure
3. Scroll to the Knowledge section:
   • Knowledge Repo URL → leave empty
   • Knowledge Local Path → click Browse, select the local directory (e.g., /Users/me/knowledge-repo)
   • Knowledge Datasets → optional, as above
4. Click Save, then restart Claude Desktop

Start it: Once configured, the knowledge MCP tools (list_datasets, get_page, lookup_index, refresh_knowledge) become available. Run /refresh-dataset after pushing changes to your knowledge repo to pull the latest.

Verifying integrations

After install, run any of these in a Claude conversation to confirm each server is up:

• Athena: "list athena tables"
• Superset: "list superset dashboards"
• Knowledge: "list datasets" or /business

If a server fails to start, check Settings → Plugins → AI Analyst Plugin → Logs for the underlying Python error (most commonly a missing AWS profile, an unreachable Superset URL, or an invalid GitHub token).

{
  "name": "amarjeet-ai-analyst",
  "owner": {
    "name": "Amarjeet Singh",
    "email": "amarjeet.singh55@hotmail.com"
  },
  "plugins": [
    {
      "name": "ai-analyst-plugin",
      "source": "./",
      "description": "AI-powered product analytics: ask a business question, get validated findings, publication-quality charts, and a slide deck.",
      "version": "2.2.0",
      "author": {
        "name": "Amarjeet Singh"
      }
    }
  ]
}