Codicil - Elixir-Native Semantic Code Search

Overview

Semantic code search and analysis for Elixir projects via MCP (Model Context Protocol). Uses compiler-level analysis for deep code understanding without external dependencies.

Source: https://github.com/E-xyza/codicil

When to Use

• Finding functions by behavior (not just name)
• Dependency analysis (function call graphs, module relationships)
• Code relationship understanding
• Elixir/BEAM native code analysis
• Compiler-level indexing (automatic during compilation)

Key Capabilities

• ✅ Semantic function search (find_similar_functions)
• ✅ Dependency analysis:
  • list_function_callers
  • list_function_callees
  • list_module_dependencies
• ✅ Module analysis (list_module_dependencies)
• ✅ Function source retrieval (get_function_source_code)
• ✅ Compiler-level indexing (automatic via tracers)
• ✅ Multi-LLM support (Anthropic, OpenAI, Cohere, Google, Grok)
• ✅ Local SQLite + vector search
• ✅ No external dependencies (uses local SQLite)

Quick Start

# mix.exs
{:codicil, "~> 0.7", only: [:dev, :test]}

# Initialize database
mix codicil.setup

# Configure environment
export CODICIL_LLM_PROVIDER=openai
export OPENAI_API_KEY=your_key

# Enable compiler tracer in mix.exs
defp elixirc_options(:prod), do: []
defp elixirc_options(_env), do: [tracers: [Codicil.Tracer]]

# Add to Phoenix router (if using Phoenix)
if Code.ensure_loaded?(Codicil) do
  plug Codicil.Plug
end

MCP Tools

find_similar_functions

Find functions by semantic description using vector similarity search.

{
  "description": "functions that validate user input",
  "limit": 10
}

list_function_callers

Find what calls a specific function (useful for debugging and impact analysis).

{
  "moduleName": "MyApp.User",
  "functionName": "create",
  "arity": 1
}

list_function_callees

Find what a function calls (useful for debugging execution paths).

{
  "moduleName": "MyApp.Orders",
  "functionName": "process",
  "arity": 1
}

list_module_dependencies

Analyze module dependencies (imports, aliases, uses, requires, runtime calls).

{
  "moduleName": "MyApp.Accounts"
}

get_function_source_code

Get complete function source code with module directives and location. Use this instead of grep or file reading.

{
  "moduleName": "MyApp.User",
  "functionName": "create",
  "arity": 1
}

Integration with ai-rules

• Plan Mode: Use Codicil for deep code understanding during architecture planning
• Build Mode: Use Codicil for dependency analysis during refactoring
• Review Mode: Use Codicil for cross-reference code patterns and consistency checking

Agent Roles

• Architect: Analyze codebase structure and dependencies using Codicil
• Orchestrator: Use Codicil for dependency impact analysis before refactoring
• Reviewer: Verify no N+1 queries and proper module structure

Tooling Integration

• Use Codicil as primary semantic search tool (Elixir-native)
• Use Codicil with jido_ai for reasoning + code understanding
• Use Codicil with swarm_ex for agent coordination + code analysis
• Use Codicil with anubis_mcp for MCP server + code understanding

Advantages Over mgrep/Serena

✅ Elixir-native (vs mgrep's Node.js) ✅ Compiler-level (vs external search tools) ✅ Multi-LLM support (flexible provider choice) ✅ No embeddings required (uses local SQLite + tracers) ✅ Local-first (runs entirely on your machine)

Workflow Example

Architecture Planning (Plan Mode)

defmodule MyApp.Architecture do
  alias Codicil.MCP.Tools

  def analyze_dependencies do
    {:ok, deps} = Codicil.MCP.Tools.ListModuleDependencies.call("MyApp.Accounts")
    
    deps
    |> Enum.each(fn dep ->
      IO.puts("#{dep.module_name} depends on: #{dep.dependencies}")
    end)
  end
end

Refactoring Support (Build Mode)

defmodule MyApp.Refactor do
  alias Codicil.MCP.Tools

  def extract_function_usage do
    {:ok, callers} = Codicil.MCP.Tools.ListFunctionCallers.call(%{
      moduleName: "MyApp.Orders",
      functionName: "process",
      arity: 1
    })
    
    {:ok, callees} = Codicil.MCP.Tools.ListFunctionCallees.call(%{
      moduleName: "MyApp.Orders",
      functionName: "process",
      arity: 1
    })
    
    # Analyze impact
    IO.puts("Called by: #{Enum.join(Enum.map(callers, & &1.name), ", ")}")
    IO.puts("Calls: #{Enum.join(Enum.map(callees, & &1.name), ", ")}")
  end
end

Best Practices

• Use Codicil for Elixir-specific code analysis
• Leverage compiler-level indexing for deep understanding
• Analyze dependencies before refactoring (callers + callees)
• Use Codicil instead of mgrep for Elixir projects
• Combine with jido_ai for complex reasoning about code
• Verify function signatures with source code retrieval
• Monitor Codicil database size (~100MB typical)

Troubleshooting

• Verify Codicil tracer is enabled in mix.exs
• Ensure environment variables are set (CODICIL_LLM_PROVIDER, OPENAI_API_KEY)
• Check Codicil database is initialized: mix codicil.setup
• Verify Phoenix router includes Codicil.Plug (if using Phoenix)
• Test Codicil MCP tools with sample queries
• Monitor indexing performance during compilation

Resources

• HexDocs: https://hexdocs.pm/codicil
• GitHub: https://github.com/E-xyza/codicil
• Stars: 41 stars, 2 forks (proven adoption)
• License: MIT (permissive)