Symbols

A fast, polyglot source code intelligence CLI. Extract symbols, parse imports, trace dependencies, and analyze impact — all from the command line.

No language server required. No build step for your projects. Just point it at your code.

Go License: MIT MCPAmpel

Table of contents

• Why this exists
• How it saves context
• What it does
• Install
• Language support
• Usage
  • Symbol extraction
  • Import parsing
  • Dependency queries
  • Impact analysis
  • Project graph summary
  • JSON output
  • Shorthand
  • Symbol search
• MCP server
• How it works
• Project root detection
• Limitations
• License

Why this exists

symbols is for the moments when you need to understand a codebase quickly without opening 30 files first.

Common pain points it targets:

• You are about to change a file and need to know blast radius immediately.
• You are onboarding to an unfamiliar repo and need a map, not a scavenger hunt.
• You are reviewing a PR and want concrete dependency and ownership signals.
• You are using AI coding tools and need reliable, structured project context on demand.

Instead of manually reconstructing context from editor tabs, grep output, and memory, symbols gives you the structural view in one step.

How it saves context

symbols saves context in two practical ways:

1. It externalizes code structure into fast, repeatable queries (list, deps, dependents, impact, graph, search) so you do not have to rebuild mental maps every session.
2. It exposes the same model through MCP (syms mcp) so agents and tools can fetch fresh project facts directly, rather than relying on stale chat history or guessed file relationships.

Net effect:

• less re-reading
• fewer "what will this break?" surprises
• faster onboarding and safer refactors
• more useful AI assistance because context is retrieved, not improvised
• lower cost from fewer exploratory engineering cycles and reduced AI token spend on repo re-discovery

What it does

syms list server.py           # functions, classes, constants, variables
syms imports server.py        # parsed import statements
syms deps server.py           # files this file imports from
syms dependents server.py     # files that import this file
syms impact server.py         # full impact analysis (direct + transitive)
syms graph .                  # project-wide dependency summary
syms search User              # find symbols by name across a project
syms mcp                      # run as MCP server for AI tools

Install

Option 1: Build from source

git clone https://github.com/Jordan-Horner/symbols.git
cd symbols
go build -o syms .
sudo mv syms /usr/local/bin/

Requirements: Go 1.26+

Option 2: Direct installation (Linux/macOS)

# Install directly to /usr/local/bin
curl -L https://github.com/Jordan-Horner/symbols/releases/latest/download/syms-$(uname -s)-$(uname -m) -o /usr/local/bin/syms
chmod +x /usr/local/bin/syms

Option 3: Homebrew (macOS)

brew tap Jordan-Horner/tap
brew install syms

Verify installation

syms --version

Language support

Symbol extraction uses tree-sitter for full AST parsing (function signatures with parameters, classes, types, constants). Import parsing and dependency resolution use regex.

Language	Symbols	Import parsing	Dependency resolution
Python	tree-sitter (functions, classes, constants, variables)	regex	Relative + absolute imports
TypeScript	tree-sitter	regex	tsconfig.json path aliases, relative paths, index.ts
JavaScript	tree-sitter	regex	Same as TypeScript (also reads jsconfig.json)
Svelte	tree-sitter (script block)	regex	Same as TypeScript
Go	tree-sitter	regex	go.mod module prefix, package directories
Java	tree-sitter	regex	Dot-to-slash, src/main/java prefix
Kotlin	tree-sitter	regex	Same as Java + .kt
Rust	tree-sitter	regex	crate/self/super, mod.rs
C#	tree-sitter	regex	Namespace-to-path, class name fallback
PHP	tree-sitter	regex	PSR-4 conventions, require/include
C/C++	tree-sitter	—	—
Ruby	tree-sitter	—	—
Scala	tree-sitter	—	—
Bash	tree-sitter	—	—

Usage

Symbol extraction

# Single file
syms list app.py

# Multiple files
syms list src/main.go src/handlers.go

# Recursive directory scan
syms list -r src/

# JSON output (for piping to other tools)
syms list --json app.py

# Pretty JSON output (human-readable)
syms list --json --pretty app.py

# Optional: include precise symbol ranges
syms list --json --ranges app.py

# Count symbols per file
syms list --count src/

# Filter by symbol kind (repeatable or comma-separated)
syms list --filter class src/
syms list --filter class,function src/
syms list --filter class --filter function src/

Output:

### `app.py` — 245 lines

  constant VERSION  # line 1
  constant API_URL  # line 3
  variable app  # line 5
  class Application  # line 12
  def __init__(self, config)  # line 15
  async def start(self)  # line 34
  def shutdown(self)  # line 78

Import parsing

syms imports server.py

Output:

### `server.py`

  from flask import Flask, jsonify  # line 1
  from .models import User, Post  # line 2
  import os  # line 3

Dependency queries

# Direct dependencies
syms deps src/handlers.go

# Transitive (everything it depends on, recursively)
syms deps -t src/handlers.go

# Who imports this file?
syms dependents src/models.py

# Transitive dependents
syms dependents -t src/models.py

Impact analysis

syms impact src/core/utils.py

Output:

### `src/core/utils.py` — impact analysis

  Direct dependents:     8
  Transitive dependents: 23

  Direct:
    src/api/handlers.py
    src/core/auth.py
    src/core/db.py
    ...

  Indirect (transitive):
    src/api/routes.py
    src/main.py
    tests/test_auth.py
    ...

Project graph summary

syms graph .

Output:

Project dependency graph

  Files:              187
  Import edges:       562
  Unresolved imports: 43

  Most depended-on files:
    src/utils.py  (36 dependents)
    src/config.py  (33 dependents)
    src/models.py  (23 dependents)

  Heaviest importers:
    src/app.py  (28 imports)
    src/main.py  (24 imports)

  Circular dependencies (1):
    src/config.py <-> src/runner.py

JSON output

All commands support --json for machine-readable output:

syms impact --json src/utils.py | jq '.direct_dependents'
syms graph --json . | jq '.hot_spots[:5]'

# Optional: pretty-print JSON for humans
syms graph --json --pretty .

# Full edge map (file → its dependencies)
syms graph --json . | jq '.edges'

# What does a specific file depend on?
syms graph --json . | jq '.edges["src/app.py"]'

Shorthand

The list subcommand is the default — you can omit it:

# These are equivalent:
syms list app.py
syms app.py

# Flags work too:
syms -r src/ --json

Symbol search

# Find symbols by name (fuzzy: exact > prefix > contains)
syms search User

# JSON output
syms search --json handle

# Search in a specific project
syms search --root /path/to/project Config

# Search only specific symbol kinds
syms search --filter class User

# Optional: include precise symbol ranges in search results
syms search --json --ranges User

Output:

Found 3 symbols matching "User":

  class User  models.py:1
  class UserProfile  models.py:5
  function get_user(id)  api/handlers.py:12

MCP server

Run syms as an MCP server for AI tool integration (e.g. Claude Code):

syms mcp

Exposes all functionality as MCP tools over stdio (JSON-RPC 2.0):

Tool	Description
syms_list	Extract symbols from files
syms_imports	Parse import statements
syms_deps	File dependencies
syms_dependents	Reverse dependencies
syms_impact	Impact analysis
syms_search	Search symbols by name
syms_graph	Project dependency graph

syms_list and syms_search accept optional kinds: string[] arguments to filter symbol kinds. syms_list and syms_search also accept optional include_ranges: boolean for start/end line+column metadata. Tool results are returned in structuredContent (not JSON text blobs in content[].text).

Claude Code setup

After installing syms, configure it as an MCP server:

Project-level (recommended for teams):

Create .mcp.json in your project root:

{
  "mcpServers": {
    "symbols": {
      "command": "syms",
      "args": ["mcp"]
    }
  }
}

Commit this file so your team gets the symbols server automatically.

Global (all projects):

Create or edit ~/.mcp.json:

{
  "mcpServers": {
    "symbols": {
      "command": "syms",
      "args": ["mcp"]
    }
  }
}

After configuration:

1. Restart Claude Code
2. When prompted, approve the symbols MCP server
3. Claude Code will now have access to code intelligence tools in all your projects

How it works

Symbol extraction uses tree-sitter for full AST parsing. Each language has a compiled grammar (linked statically into the binary) that produces a syntax tree. The tool walks the tree to extract top-level declarations with names, kinds, line numbers, and function parameters. For Python, module-level assignments are also extracted as constants (UPPER_CASE) or variables.

Import parsing uses regex patterns tuned to each language's import syntax. This is fast and reliable for standard import forms without needing AST parsing.

Dependency resolution maps import specifiers to actual files on disk using language-specific conventions:

• Python: module dot-path to file path, relative import resolution
• Go: go.mod module name stripping, package-to-directory mapping
• Java/Kotlin: dot-to-slash convention, standard source root prefixes (src/main/java/)
• Rust: crate/self/super path resolution, mod.rs convention
• C#: namespace-to-path with progressive prefix stripping
• PHP: PSR-4 backslash-to-slash mapping, require/include path resolution

Directory scanning uses early pruning of .git, node_modules, dist, build, vendor, target, and other common non-source directories.

Project root detection

For deps, dependents, impact, and graph, the tool auto-detects the project root by walking up the directory tree looking for .git, package.json, or pyproject.toml. Override with --root:

syms deps src/app.py --root /path/to/project

Limitations

• Convention-based resolution — dependency resolution uses file path conventions, not compiler/build system integration. TypeScript/JavaScript paths from tsconfig.json/jsconfig.json are supported (including extends), but webpack/vite aliases defined outside tsconfig are not.
• File-level granularity — dependencies are traced at the file level (import graph), not at the function or symbol level. There is no call graph.
• C/C++ includes — #include parsing and header resolution are not yet implemented. Symbol extraction works, but dependency tracing does not.
• Ruby/Scala/Bash — symbol extraction works via tree-sitter, but import parsing and dependency resolution are not implemented.
• Dynamic imports — Python's importlib.import_module(), JavaScript's computed require(), and similar dynamic patterns are not detected.
• Monorepo boundaries — the tool resolves imports within a single project root. Cross-package imports in monorepos may not resolve correctly.

License

MIT