xtage
Provides Claude with persistent, semantic knowledge of a codebase across sessions by maintaining a local knowledge base of structural maps, compressed file summaries, and insights.
README
xtage
CLI + MCP server that gives Claude persistent, semantic knowledge of a codebase — across every session, without manual context-setting.
Author: ruco-ai | Date: 2026-04-26
Overview
Claude starts every session cold. xtage fixes this by maintaining a persistent knowledge base that Claude reads before touching any source files.
The knowledge base lives in ~/xtage/ and is structured in three tiers:
- Structural map —
CODEINDEX.md,REPO.md,PROJECTINSIGHTS.mdper repo - Per-file compressed summaries —
files/{slug}.mdper source file - Source files — opened only when the above tiers are insufficient
All files are generated by your local Claude, live in ~/xtage/, and stay on your machine. No external AI API. No cloud.
Installation
Claude Code (recommended):
claude mcp add xtage npx -- -y xtage
For global installation across all projects, edit ~/.claude/settings.json:
{
"mcpServers": {
"xtage": {
"command": "npx",
"args": ["-y", "@ruco-ai/xtage"]
}
}
}
Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"xtage": {
"command": "npx",
"args": ["-y", "@ruco-ai/xtage"]
}
}
}
Usage
1. Index a repo
xtage init
Indexes the current directory. Local filesystem by default — works offline, works for private repos, indexes your actual working state.
To index a public GitHub repo instead:
xtage init https://github.com/owner/repo
Claude fetches, chunks, and writes REPO.md and CODEINDEX.md to ~/xtage/{repo-name}/. A spinner shows live progress.
2. Start a session
xtage start
Loads your codebase context into Claude at the start of a session. Claude reads the knowledge base silently and answers your first question with it already loaded.
3. End a session
xtage end
Claude reflects on what was learned and writes new entries to PROJECTINSIGHTS.md or GENERALINSIGHTS.md.
4. Update after commits
xtage update
Incrementally refreshes CODEINDEX.md based on what changed since the last index. Much faster than a full re-init.
Install as a git post-commit hook to run automatically:
echo 'xtage update' >> .git/hooks/post-commit
chmod +x .git/hooks/post-commit
5. Check index freshness
xtage status
Shows when the index was last updated and whether it's stale.
6. Register a local directory
xtage link
Registers the current directory in the local repo registry (~/xtage/registry.json), so xtage can resolve which indexed repo corresponds to the current working directory.
MCP Tools
Fetch tools
| Tool | Description |
|---|---|
fetch_repo_data |
Fetch a public GitHub repo in token-sized chunks |
fetch_local_repo_data |
Fetch files from a local directory in token-sized chunks |
get_file |
Fetch a single file from a GitHub repo |
check_for_updates |
Check if CODEINDEX.md is stale vs latest GitHub commit |
Write tools
| Tool | Description |
|---|---|
write_repo_md |
Write REPO.md (architecture, stack, workflow) |
write_codeindex |
Write CODEINDEX.md (file-by-file semantic map) |
write_project_insights |
Write PROJECTINSIGHTS.md (design decisions, gaps) |
write_general_insights |
Write global GENERALINSIGHTS.md (cross-repo patterns) |
write_file_chunk |
Write a per-file prose summary to files/{slug}.md |
Read tools
| Tool | Description |
|---|---|
read_codeindex |
Read CODEINDEX.md |
read_repo_md |
Read REPO.md |
read_project_insights |
Read PROJECTINSIGHTS.md |
read_file_chunk |
Read a per-file summary from files/{slug}.md |
list_file_chunks |
List available file chunk summaries for a repo |
get_session_stats |
Token savings stats for the current session |
Registry tools
| Tool | Description |
|---|---|
register_repo |
Map a local path → repo name in ~/xtage/registry.json |
lookup_repo |
Resolve local path → repo name (or list all registered repos) |
Memory sync
| Tool | Description |
|---|---|
sync_from_claude_memory |
Merge Claude auto-memory files (~/.claude/projects/.../memory/) into PROJECTINSIGHTS.md. Wired automatically via a Claude Code PostToolUse hook — fires whenever a memory file is written. Silently skips repos not registered in xtage. |
Storage Layout
~/xtage/
registry.json ← local path → repo name mappings
GENERALINSIGHTS.md ← cross-repo patterns and standing instructions
{repo-name}/
REPO.md ← architecture, stack, workflow
CODEINDEX.md ← file-by-file semantic map
PROJECTINSIGHTS.md ← design decisions, known gaps
files/
{slug}.md ← per-file compressed prose summary
Configuration
| Variable | Default | Description |
|---|---|---|
XTAGE_NO_TELEMETRY |
unset | Set to 1 to disable anonymous telemetry |
Add a .xtage-ignore file to your repo root to exclude paths (same format as .gitignore).
Default ignores: .env, *.pem, *.key, secrets/, node_modules/, dist/, *.lock.
Project Structure
xtage/
src/
cli.ts ← CLI entry point (commands are thin proxies for claude -p)
server.ts ← MCP server setup
tools.ts ← MCP tool definitions
resources.ts ← MCP resources (xtage:// URIs)
prompts.ts ← MCP prompts (repo-init, repo-update, session-start, session-end)
chunker.ts ← token-aware file chunker
github.ts ← GitHub REST API fetcher
local.ts ← local filesystem repo reader
ignore.ts ← ignore chain + secretlint scanning
store.ts ← read/write ~/xtage/ files + registry
progress.ts ← spinner progress state (CLI ↔ MCP via env var)
sessions.ts ← session token tracking
telemetry.ts ← anonymous usage telemetry
hook.ts ← git diff parser (installGitHook removed; parseGitDiff + getGitDiff only)
tokens.ts ← local token counter
version.ts
templates/ ← mdblu templates for xtage index files
Telemetry
xtage collects anonymous usage data to understand how the tool is used. It is opt-in — you are asked on first run, and the default is off. Set XTAGE_NO_TELEMETRY=1 to disable permanently without being asked.
What is collected: event type (init, update, error, general_insights_write), timestamp, repo URL, file/chunk counts, word count of insights written. No source code, no file contents, no personal data.
Preference is stored at ~/.xtage/telemetry.json.
Release Notes
0.4.0
- New:
sync_from_claude_memoryMCP tool — reads Claude auto-memory files and merges them intoPROJECTINSIGHTS.md. Wire it with a Claude CodePostToolUsehook so every memory write propagates to xtage automatically. - Removed: dead
init.tsandupdate.tsmodules (logic had already moved tocli.tsviarunClaude); removedinstallGitHookfromhook.ts. - Fixed:
sync_from_claude_memoryadded toXTAGE_TOOLSsoensureXtagePermissions()grants it onxtage init.
0.3.x
- Session token tracking and
get_session_statstool - Per-file chunk summaries (
write_file_chunk,read_file_chunk,list_file_chunks) - Open-source release: LICENSE, telemetry disclosure,
.npmignore
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.