Repowise MCP server
{
"mcpServers": {
"repowise": {
"command": "uvx",
"args": [
"repowise"
]
}
}
}{
"mcpServers": {
"repowise": {
"command": "uvx",
"args": [
"repowise"
]
}
}
}{
"mcpServers": {
"repowise": {
"command": "uvx",
"args": [
"repowise"
]
}
}
}{
"servers": {
"repowise": {
"command": "uvx",
"args": [
"repowise"
]
}
}
}{
"mcpServers": {
"repowise": {
"command": "uvx",
"args": [
"repowise"
]
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsoninstall uvx repowiseversion 0.17.1
What it does
Repowise indexes codebases to provide dependency graphs, git history, tests, documentation, and architectural decisions to AI agents and developers. It exposes this indexed information through MCP tools, enabling agents to understand repositories without redundant analysis and developers to assess change impact and code health across single or multiple repositories.
- Provide AI agents with cited, task-shaped context about code structure and decisions to reduce tool calls and exploration overhead.
- Analyze pull request changes to identify symbol-level blast radius, determine which tests must run, and detect breaking API contracts before merge.
- Surface defect-prone files and generate concrete refactoring plans informed by code health scores and dependency graphs.
Tools this server exposes
| Tool | Does |
|---|---|
| graph | Query dependency and call graphs across the indexed codebase. |
| change_impact | Analyze symbol-level blast radius and affected tests for a given diff. |
| code_health | Retrieve defect risk, maintainability, and performance scores with concrete refactoring suggestions. |
| documentation | Access generated and indexed documentation kept current with the codebase. |
| decisions | Retrieve architectural decisions and context explaining why code exists. |
| tests | Query test coverage and determine which tests guard specific code paths. |
| dead_code | Identify unused symbols and dead code paths. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.