Skip to main content
Kasava exposes 8 live-data tools (plus 3 AI agents and 2 workflows) via the Model Context Protocol (MCP). Any MCP-compatible client can connect to search code, query commits, search customer signals, load plans, and more. Cacheable data (product graph, work breakdowns, ownership, rationale) lives in the local .kasava/ artifact maintained by the Kasava CLI — it’s not in MCP because reading filesystem-cached data costs zero tokens at session start while live MCP schemas don’t.

Connection Details

Prerequisites

  1. A Kasava account with at least one indexed repository
  2. An API key from Settings > API Keys

Client Setup

Claude Code

The easiest option is to install the Kasava plugin, which configures MCP automatically. For manual setup, add to your project’s .mcp.json:

Cursor

Add to your Cursor MCP settings (~/.cursor/mcp.json or project-level):
See the Cursor-specific guide for detailed instructions.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

Windsurf

Add to your Windsurf MCP configuration:

Available Tools

Once connected, your AI agent has access to 8 tools, 3 AI agents, and 2 workflows. See the MCP Tool Reference for the complete list. Key capabilities:
  • Code intelligence (code-tool): Symbol search, definitions, usages, dependencies, semantic search, comments, dead code, code smells
  • Commit analysis (commit-tool): Semantic classification, hotspots, changelog
  • Unified search (search-tool): One-call multi-corpus search with plan staleness flags
  • Customer signals (document-search-tool): Search uploads, plans, signals, product docs
  • Competitive intel (competitor-tool): Live scrapings, change detection, market trends
  • Graph traversal (entity-analysis-tool): Feature / layer / symbol relationships
  • Plan specs (spec-export-tool): Agent-prompt format
  • Decision recording (record-decision-tool): Write decisions against the bound product
  • AI agents: ask_codeIntelligenceAgent, ask_productInsightsAgent, ask_architectureInvestigationAgent
  • Workflows: run_bugAnalysisWorkflow, run_deepResearchWorkflow
What’s not in MCP — and why:
  • Product graph queries → cached in .kasava/graph.json after kasava sync (zero token cost)
  • Work breakdowns.kasava/work-items.json
  • Plan markdown.kasava/plans/<slug>.md
  • File ownership / rationalekasava ownership <file> and kasava rationale <file> CLI commands
  • GitHub / Linear / Jira / Asana writes → web app or gh CLI

Verify Connection

Test your connection with curl:
You should see SSE events streaming. Press Ctrl+C to disconnect.

Troubleshooting

Check that your API key is valid and hasn’t expired. Try creating a new key at Settings > API Keys.
Ensure your Authorization header uses the Bearer prefix: Authorization: Bearer kasava_live_...
After connecting, your MCP client needs to discover tools. In Claude Code, try asking “What Kasava tools are available?” In Cursor, check the MCP panel.