Docs/AI & Automation/Claude Code & MCP Integration

Claude Code & MCP Integration

senSEi includes a Model Context Protocol (MCP) server. When you connect Claude Code to it, your laptop becomes the AI compute layer — all analyses run through the latest Claude models on your machine. The MCP server handles data access; Claude Code handles reasoning.

Why this matters

The senSEi LiteLLM proxy is IP-restricted to the Okta network. By routing AI calls through Claude Code on your laptop, you get:

Access to the latest Claude models (Opus 4.8, Sonnet 4.6)
Full deal context in every analysis — Slack messages, email threads, Google Drive docs, Lucid diagrams, Gong summaries, attachments, and transcripts all included
An agent compute layer you control — analyses only run when you're in a session
45 MCP tools covering every read and write operation in the app

Setup

Go to Settings → MCP in senSEi.
Click Generate API Key to create a scoped key for your user.
Copy the server URL: https://sensei.oktademo.app/api/mcp

Add the server to your Claude Code MCP config at ~/.claude/settings.json:

{
  "mcpServers": {
    "sensei": {
      "type": "http",
      "url": "https://sensei.oktademo.app/api/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

Restart Claude Code. The sensei tools appear automatically.

Available tools (45)

Tools are grouped into 7 categories:

Accounts & Opportunities — search, get, create, update accounts and opportunities
Meetings — create, get, update meetings; search transcripts
Contacts — create and update contacts on accounts
Intelligence — add Slack messages, email threads, Drive docs, Lucid diagrams, Gong data; get insights
Analysis — get analysis context, write results, run analyses, get health score
Pipeline & Boards — list pipeline, create and update cards
Agent Automation — get and drain the analysis queue, get and update agent preferences

Analysis queue and dirty flags

Write tools that affect deal context — adding Slack messages, saving transcripts, updating COM fields — automatically set dirty flags on the affected opportunity. These accumulate into an analysis queue.

Claude Code can drain this queue in two ways:

Automatic — If Auto-run on session start is enabled in Settings → Agent Automation, Claude Code checks the queue at session start and processes it.
Manual — Run /analyze-queue in Claude Code to see the pending queue and process it. Run /run-analysis <nodeId> <type> for a single on-demand analysis.

Analysis types

com — Command of the Message synthesis (7 Okta Force Management fields)
three_whys — Why Anything, Why Okta, Why Now narrative
state — Current State and Future State architecture diagrams
presales — Okta SFDC presales field population + leadership-quality meeting notes
meeting — Meeting summary, action items, key quotes

Security

MCP API keys are scoped to your user and org. They can't access other users' data. Revoke any key from Settings → MCP at any time. Keys are stored encrypted at rest.

← Back to docs Open app →