lidx

Code indexer with MCP server for LLM-assisted code navigation.

lidx indexes your codebase into a symbol graph with cross-language references, then serves queries over MCP so AI coding assistants can navigate your code intelligently — finding callers, tracing data flow across service boundaries, and understanding impact before making changes.

Install

curl -fsSL https://raw.githubusercontent.com/JacobSoderblom/lidx/main/install.sh | bash

Downloads a prebuilt binary for your platform (macOS/Linux, x64/ARM64) to ~/.local/bin.

Setup

Add .lidx to your repo's .gitignore:

echo ".lidx" >> .gitignore

Add lidx to your repo's .mcp.json:

{
  "mcpServers": {
    "lidx": {
      "command": "lidx",
      "args": ["mcp-serve", "--repo", "."]
    }
  }
}

That's it. Your AI assistant can now query your codebase through lidx.

What it does

lidx builds a symbol graph from your code and exposes it through a single MCP tool (lidx_query) with ~30 query methods. The graph captures:

• Symbols — functions, classes, methods, modules, routes, SQL tables
• Edges — CALLS, IMPORTS, INHERITS, TYPE_REF, HTTP_CALL/HTTP_ROUTE, RPC_CALL/RPC_IMPL, CHANNEL_PUBLISH/CHANNEL_SUBSCRIBE
• Cross-language references — C# calling SQL stored procedures, C# referencing Python classes, gRPC service implementations across languages

Supported languages

Python, C#, TypeScript, JavaScript, Rust, Go, Lua, SQL, PostgreSQL (PL/pgSQL), Proto, Markdown

Key capabilities

Navigation — find symbols, jump to definitions, explore neighbors, trace call chains

Impact analysis — multi-layer analysis (direct graph + test coverage + git co-change history) with confidence scoring. Understands what changes when you modify a symbol.

Cross-service tracing — automatically bridges service boundaries via gRPC, HTTP routes, and message bus channels (Azure Service Bus, RabbitMQ). Trace a request from API endpoint through message queue to background worker across languages.

Context assembly — gather_context assembles LLM-ready context from symbols, files, and search queries within a byte budget. One call gives your AI assistant exactly the code it needs.

Incremental indexing — watches for file changes and re-indexes automatically. No manual reindex needed during development.

Query methods

All queries go through the lidx_query MCP tool with a method and optional params.

Category	Methods
Discovery	help, list_methods, list_languages, repo_overview, repo_insights
Search	find_symbol, suggest_qualnames, search_text, search_rg, grep
Navigation	open_symbol, open_file, neighbors, subgraph, references
Analysis	gather_context, analyze_impact, analyze_diff, trace_flow
Routes	route_refs, flow_status
Index	changed_files, index_status, reindex
Quality	top_complexity, duplicate_groups, diagnostics_run, diagnostics_list, diagnostics_summary

Every result includes next_hops — pre-computed follow-up queries so your AI assistant knows what to explore next.

Example prompts

These are things you can say to your AI assistant once lidx is connected. The assistant will automatically call the right lidx queries behind the scenes.

Discovery

• "Give me an overview of this repo"
• "What languages are in this project?"
• "Show me the most complex functions in the codebase"
• "Are there any duplicated code blocks?"

Finding code

• "Find the UserService class"
• "Search for anything related to authentication"
• "Find all gRPC service definitions"
• "Search for TODO comments across the repo"

Understanding code

• "Explain what the OrderProcessor class does — its callers, callees, and dependencies"
• "What does the register_user function call?"
• "Show me all the callers of validate_email"
• "What imports does this module depend on?"

Impact analysis

• "If I change the User model, what else is affected?"
• "What's the blast radius of modifying the payment processing function?"
• "Show me the impact of this diff" (with a git diff or file changes)
• "What tests cover the authentication module?"

Tracing across services

• "Trace the flow starting from the /api/orders HTTP endpoint"
• "How does a message published to the order-created topic get processed?"
• "Show me the full request path from the API gateway through gRPC to the database"
• "What services are connected to the UserUpdated message bus channel?"

Navigation

• "Show me the neighbors of the PaymentGateway class"
• "Expand the call graph around processOrder, 3 levels deep"
• "What are the HTTP routes in this project and who calls them?"
• "Show me all cross-language references between C# and SQL"

Code quality

• "Run diagnostics on the Python code"
• "What are the most complex functions? Any above complexity 10?"
• "Find dead symbols — functions that are never called"
• "Are there any unused imports?"

Context for tasks

• "Gather context about the authentication system — I need to add OAuth support"
• "Pull together everything related to the order processing pipeline"
• "I need to understand the database schema and all functions that touch the users table"

CLI usage

# Index a repo
lidx reindex --repo /path/to/repo

# Run MCP server (used by .mcp.json)
lidx mcp-serve --repo .

# Run JSONL RPC server
lidx serve --repo .

# One-off query
lidx request --method find_symbol --params '{"query":"MyClass"}'

Configuration

Variable	Default	Description
LIDX_SEARCH_TIMEOUT_SECS	30	Ripgrep search timeout
LIDX_PATTERN_MAX_LENGTH	10000	Max search pattern length in bytes
LIDX_POOL_SIZE	10	SQLite read connection pool size
LIDX_POOL_MIN_IDLE	2	Minimum idle connections

Ignore rules

By default lidx respects .gitignore. Use --no-ignore with reindex, serve, or mcp-serve to include ignored files. For searches, pass no_ignore: true to search_text, grep, or search_rg.

License

MIT