MCP Tool Reference

When connected via MCP, your AI agent has access to 8 tools, 3 server-side AI agents, and 2 multi-step workflows. The toolset was right-sized in May 2026 (cut from 23 — see ADR-0015) so cacheable data — product graph, work breakdowns, ownership, rationale, health — lives in your local .kasava/ artifact instead, refreshed by kasava sync. Live data (semantic search, customer signals, agents, write surfaces) stays in MCP.

Authentication

All MCP connections require a Kasava API key. Create one at Settings → API Keys. The MCP server endpoint is:

https://api.kasava.dev/v1/mcp/sse

Example: Connecting from Claude Code

The Kasava plugin configures this automatically. To do it manually:

{
  "mcpServers": {
    "kasava": {
      "url": "https://api.kasava.dev/v1/mcp/sse",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Example: Connecting from Cursor

{
  "mcpServers": {
    "kasava": {
      "serverUrl": "https://api.kasava.dev/v1/mcp/sse",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Tools (8)

code-tool

Search and navigate codebases using Kasava’s symbol graph.

Action	Description	Key Parameters
`search_symbols`	Find symbols by name or pattern	`query`, `repositoryId`, `symbolTypes`, `limit`
`get_definition`	Get full source of a symbol	`symbolName`, `repositoryId`
`find_usages`	Find all references to a symbol	`symbolName`, `repositoryId`, `limit`
`find_dependencies`	Trace symbol dependencies	`symbolName`, `repositoryId`, `maxDepth`
`get_file` / `get_files`	Read file content	`filePath`, `repositoryId`, `branch`
`get_related_files`	Walk dependency graph from a file	`filePath`, `repositoryId`
`list_files`	List files in a path	`repositoryId`, `path`
`semantic_search`	Natural language code search	`query`, `repositoryId`, `limit`
`search_comments`	Search comments and docstrings	`query`, `repositoryId`
`search_tech_debt`	Find TODO/FIXME/HACK markers	`repositoryId`, `limit`
`find_dead_code`	Find unused exported symbols	`repositoryId`, `limit`
`find_code_smells`	Detect code quality issues	`repositoryId`, `limit`

commit-tool

Query commits with semantic analysis.

Action	Description	Key Parameters
`list`	List commits with filters	`repositoryId`, `semanticType`, `impact`, `author`, `dateRange`
`get`	Get commit details	`sha`, `repositoryId`
`hotspots`	Find frequently changed files	`repositoryId`, `limit`
`changelog`	Generate markdown changelog	`repositoryId`, `dateRange`

search-tool

Unified context-aware search across code, issues, plans, docs, competitors, and comments. Carries plan staleness flags and code references inline.

Action	Description	Key Parameters
`search`	Single-call multi-corpus search	`query`, `productId`, `corpora`, `limit`

document-search-tool

Search uploads, plans, customer signals, and product documentation.

Action	Description	Key Parameters
`search_uploads`	Search uploaded documents	`query`, `productId`, `limit`
`search_plans`	Search plan content	`query`, `productId`, `limit`
`search_product_docs`	Search indexed documentation	`query`, `productId`, `limit`, `minSimilarity`
`search_signals`	Search customer signals	`query`, `signalType`, `urgency`, `productId`

competitor-tool

Live competitive intelligence and market analysis.

Action	Description
`get_intel`	Structured competitor analysis
`search_docs`	Search competitor documentation
`search_trends`	Market trend analysis
`get_competitive_intel`	Full competitive intelligence
`get_gong_signals`	Competitor mentions from sales calls

entity-analysis-tool

Analyze relationships between code entities (graph traversal).

Parameter	Type	Description
`entityType`	string	Type of entity to analyze
`entityId`	string	Entity identifier
`includeRelated`	boolean	Include related entities
`includeImpact`	boolean	Include impact analysis
`includeGraph`	boolean	Include dependency graph
`includeExplanation`	boolean	Include AI explanation

spec-export-tool

Export plans as agent-consumable specifications. Returns the agent-prompt format (condensed, clipboard-ready). For full markdown specs, read .kasava/plans/<slug>.md after kasava sync.

Parameter	Type	Description
`planId`	string	Plan to export
`format`	`markdown` \| `json`	Output format
`includeCodeContext`	boolean	Include relevant code references

record-decision-tool

Record an agent decision against a plan (“done, deviated because X”). Closes the plan-to-code feedback loop. The only write surface exposed via MCP.

Parameter	Type	Description
`planId`	string	Plan this decision relates to
`decisionType`	`architecture` \| `technology` \| `feature` \| `process` \| `scope`
`summary`	string	Short decision summary
`rationale`	string	Why this decision was made

Agents (3)

These agents orchestrate multiple tools for complex queries.

Agent	Invoke with	Purpose
Code Intelligence	`ask_codeIntelligenceAgent`	Deep code analysis, symbol tracing across many files
Product Insights	`ask_productInsightsAgent`	Product health, architecture overview, sprint analytics
Architecture Investigation	`ask_architectureInvestigationAgent`	Architecture analysis and recommendations

Workflows (2)

Workflow	Invoke with	Purpose
Bug Analysis	`run_bugAnalysisWorkflow`	Multimodal bug analysis with code search + AI synthesis
Deep Research	`run_deepResearchWorkflow`	Multi-step research with web search

Usage Examples

Here are natural language prompts and the tools they invoke:

Prompt	Tools used
”How does checkout work in our app?”	`code-tool.search_symbols` → `get_definition` → `find_dependencies`
”What competitors are doing with AI features?”	`competitor-tool.get_competitive_intel`
”Show me recent high-impact commits”	`commit-tool.list` (impact: high)
“Find all customer reports about login”	`document-search-tool.search_signals` (signalType: bug_report)
“Research how other companies handle rate limiting”	`run_deepResearchWorkflow`
”Export our auth plan as a starter prompt for an agent”	`spec-export-tool` (format: markdown)
“I’m done with the rate-limit task — record that we used a sliding window”	`record-decision-tool` (type: technology)

What’s not in MCP (and where it lives)

These data sources moved out of the MCP toolset in May 2026 — they live in your local .kasava/ artifact (refreshed by kasava sync) or are accessed via CLI / web app.

Concept	Where it lives now
Product graph (features, layers, tech stack)	`.kasava/graph.json`, read via the `kasava-context` skill or any tool
Work breakdowns	`.kasava/work-items.json`
Plan agent-spec markdown	`.kasava/plans/<slug>.md` (also via `kasava plan` and `spec-export-tool`)
File ownership	`kasava ownership <file>` (CLI)
File rationale (TODO/FIXME + referencing docs)	`kasava rationale <file>` (CLI)
Repository / integration health	Web app dashboards
Linear / Jira / Asana writes	Web app — Plans → Export to Platform
GitHub issue / PR operations	The `gh` CLI; or Linear/Jira/Asana via web export
Long-running deep research	`run_deepResearchWorkflow` (workflow) or web app Research surface
Report generation	Web app dashboards
Code diff rendering	`git diff`

Getting Started

Concepts

Core Features

Agent Capabilities

Integrations

Developer Tools

Settings

MCP Tool Reference

Authentication

Example: Connecting from Claude Code

Example: Connecting from Cursor

Tools (8)

code-tool

commit-tool

search-tool

document-search-tool

competitor-tool

entity-analysis-tool

spec-export-tool

record-decision-tool

Agents (3)

Workflows (2)

Usage Examples

What’s not in MCP (and where it lives)

​Authentication

​Example: Connecting from Claude Code

​Example: Connecting from Cursor

​Tools (8)

​code-tool

​commit-tool

​search-tool

​document-search-tool

​competitor-tool

​entity-analysis-tool

​spec-export-tool

​record-decision-tool

​Agents (3)

​Workflows (2)

​Usage Examples

​What’s not in MCP (and where it lives)

Authentication

Example: Connecting from Claude Code

Example: Connecting from Cursor

Tools (8)

code-tool

commit-tool

search-tool

document-search-tool

competitor-tool

entity-analysis-tool

spec-export-tool

record-decision-tool

Agents (3)

Workflows (2)

Usage Examples

What’s not in MCP (and where it lives)