Terrain
Terrain indexes entire codebases into a queryable knowledge graph, providing AI coding assistants with precise function signatures, call graphs, and semantic search across all code.
README
Terrain
English | Chinese / CN
Give your AI coding assistant a complete map of any codebase — function signatures, call graphs, and semantic search across every line of code.
The Problem
You drop a 500,000-line codebase in front of Claude Code. It reads what it can see. It guesses what it can't. You get answers that are almost right.
Terrain indexes the entire codebase once, then gives your AI a precise, queryable knowledge graph — so it stops guessing.
What This Looks Like
Ask Claude Code about an unfamiliar codebase:
"How does the authentication token get refreshed?"
Without Terrain, the AI skims files and makes educated guesses — possibly missing the real implementation buried three call levels deep.
With Terrain:
find_api("authentication token refresh")
→ refresh_access_token() in auth/token_manager.c:187
Signature: int refresh_access_token(TokenCtx *ctx, const char *refresh_token)
Called by: session_heartbeat() → event_loop_tick() → main()
Calls: http_post(), parse_jwt(), update_session_store()
Precise. Complete. Instant.
Full Installation
Install the npm package
The npm package provides the CLI wrapper and MCP server launcher:
npm install -g terrain-ai@latest
Install the Python package (PyPI)
The Python package provides the core indexing engine, graph database, and all language parsers.
Core installation (includes C/C++, Python, JavaScript/TypeScript grammars):
pip install terrain-ai
Quick Start — Agent Install (Recommended)
Already using an AI agent like Claude Code, opencode, or codex? Paste this into your agent chat:
Please follow the installation instructions at:
https://raw.githubusercontent.com/JeremyJiao01/Terrain-AI/main/AGENT_INSTALL.md
Your agent will handle everything: Python 3.11 check, package install, API key setup, and MCP registration. After the first run, you can re-trigger by saying "install terrain" in any session.
Manual Install (Alternative)
npx terrain-ai@latest --setup
The setup wizard installs the Python package, configures your LLM and embedding provider, and registers Terrain as a global MCP server for supported clients. One command.
Supported clients (auto-detected):
- Claude Code — MCP registered via
claude mcp add; slash commands installed to~/.claude/commands/ - opencode — MCP registered by editing
~/.config/opencode/opencode.json(respects$XDG_CONFIG_HOME); slash commands installed to~/.config/opencode/command/
Index a Codebase
terrain index /path/to/your/repo
Takes a few minutes the first time. Incremental updates after that:
terrain index -i # git-diff based, fast
What You Can Ask
| You want to know... | What to ask |
|---|---|
| Where does X get initialized? | "find where X is initialized" |
| What calls this function? | "find callers of function_name" |
| How does feature Y work end-to-end? | "trace the call chain for Y" |
| What functions handle Z? | "find Z handler" |
Supported Languages
C/C++, Python, JavaScript/TypeScript, Rust, Go, Java, Scala, C#, PHP, Lua
Reference
Uninstall
npx terrain-ai@latest --uninstall
Removes: Claude MCP registration, opencode MCP registration, slash commands from both clients, Python package, workspace data.
CLI Tool (terrain)
Workspace
terrain status # Show active repository, workspace, LLM & embedding info
terrain list # List all indexed repositories
terrain repo # Interactively switch active repository
terrain config # Interactive configuration wizard (LLM, embedding, workspace)
terrain link <path> # Link a local repo to shared pre-built artifacts
terrain link <path> --db x # Link to a specific artifact directory
Indexing
terrain index # Index current directory (graph → api-docs → embeddings)
terrain index /path/to/repo # Index a specific path
terrain index -i # Incremental update (git-diff based, fast)
terrain index --no-embed # Skip embedding generation
terrain index --no-wiki # Skip wiki generation only
Rebuild & Clean
terrain rebuild # Rebuild all steps for active repository
terrain rebuild --step graph # Rebuild only the graph
terrain rebuild --step api # Rebuild only API docs
terrain rebuild --step embed # Rebuild only embeddings
terrain rebuild --step wiki # Rebuild only wiki
terrain clean # Remove indexed data (interactive)
terrain clean repo_name # Remove specific repository
terrain clean --all # Remove all indexed repositories
Low-Level Commands
terrain scan /path # Scan repo and build knowledge graph
--backend kuzu|memgraph|memory
--db-path ./graph.db
--exclude "vendor,build"
--language "c,python"
--clean # Clean DB before scanning
-o graph.json # Export graph to JSON
terrain query "MATCH (f:Function) RETURN f.name LIMIT 10"
--format table|json
terrain export /path -o graph.json
--build # Build graph before exporting
terrain stats # Show graph statistics (nodes, relationships)
Global Flags
terrain --version # Show version
terrain -v ... # Verbose/debug output
terrain --help # Show help
MCP Tools
Core workflow for AI agents: initialize_repository → find_api → get_api_doc
Repository Management
| Tool | Description |
|---|---|
initialize_repository |
Index a repo: graph + API docs + embeddings |
get_repository_info |
Active repo stats (node/relationship counts, service status) |
list_repositories |
All indexed repos with pipeline completion status |
switch_repository |
Switch active repo for queries |
link_repository |
Reuse existing index for a different repo path (no re-indexing) |
Code Search & Documentation
| Tool | Description |
|---|---|
find_api |
Hybrid semantic + keyword search with API doc (primary search tool) |
list_api_docs |
Browse L1 module index or L2 module details |
get_api_doc |
L3 function detail: signature, call tree, usage examples, source |
generate_api_docs |
Generate/update API docs (full / resume / enhance) |
Call Graph Analysis
| Tool | Description |
|---|---|
find_callers |
Find all functions that call a specific function (no LLM required) |
trace_call_chain |
BFS upward call chain trace with entry point discovery |
Configuration & Maintenance
| Tool | Description |
|---|---|
rebuild_embeddings |
Build or rebuild vector embeddings |
Pipeline
| Step | What | Input | Output |
|---|---|---|---|
| 1. graph-build | Tree-sitter AST parsing | Source code | Kuzu graph database |
| 2. api-doc-gen | Query graph, render docs | Graph | 3-level Markdown (index / module / function) |
| 2b. desc-gen | LLM generates descriptions | Functions without docstrings | Descriptions in L3 Markdown |
| 3. embed-gen | Vectorize function docs | L3 Markdown files | Vector store (pickle) |
initialize_repository -> Steps 1-3 (full pipeline)
build_graph -> Step 1 only
generate_api_docs -> Step 2 + 2b (modes: full / resume / enhance)
rebuild_embeddings -> Step 3
generate_wiki -> Separate (not in main pipeline)
API Documentation Format
Generated docs are optimized for both AI agent reading and vector retrieval.
L3 Function Detail (embedding unit)
# parse_btype
> Parse base type declaration including struct/union/enum specifiers.
- Signature: `int parse_btype(CType *type, AttributeDef *ad, int ignore_label)`
- Return: `int`
- Visibility: static | Header: tccgen.h
- Location: tccgen.c:139-280
- Module: tinycc.tccgen --C code generator
## Call Tree
parse_btype
|-- expr_const [static]
|-- parse_btype_qualify [static]
|-- struct_decl [static]
| |-- expect
| `-- next
`-- parse_attribute [static]
## Called by (5)
- type_decl (tinycc.tccgen) -> tccgen.c:1200
- post_type (tinycc.tccgen) -> tccgen.c:1350
C/C++ Specific Features
- Extracts
//and/* */comments above functions as descriptions - Struct/union/enum members displayed with types
- Macro definitions in dedicated section
- Static/public/extern visibility classification
- Memory ownership inference from signatures
- Header/implementation file split
- Cross-file function call resolution via
#includeheader mapping - Function pointer tracking and indirect call resolution
- GB2312/GBK encoding support for source files
Supported Languages (detail)
| Language | Functions | Classes/Structs | Calls | Imports | Types |
|---|---|---|---|---|---|
| C / C++ | Yes | struct, union, enum, typedef, macro | Yes | #include | Yes |
| Python | Yes | Yes | Yes | Yes | - |
| JavaScript / TypeScript | Yes | Yes | Yes | Yes | - |
| Rust | Yes | struct, enum, trait, impl | Yes | Yes | - |
| Go | Yes | struct, interface | Yes | Yes | - |
| Java | Yes | class, interface, enum | Yes | Yes | - |
| Scala | Yes | class, object | Yes | Yes | - |
| C# | Yes | class, namespace | Yes | - | - |
| PHP | Yes | class | Yes | - | - |
| Lua | Yes | - | Yes | - | - |
Graph Schema
Nodes: Project, Package, Module, File, Folder, Class, Function, Method, Type, Enum, Union
Relationships: CONTAINS_*, DEFINES, DEFINES_METHOD, CALLS, INHERITS, IMPLEMENTS, IMPORTS, OVERRIDES
Properties: qualified_name (PK), name, path, start_line, end_line, signature, return_type, visibility, parameters, kind, docstring
Architecture
The project follows a 5-layer harness architecture:
L4 entrypoints/ MCP server, CLI
L3 domains/upper/ apidoc, rag, guidance, calltrace
L2 domains/core/ graph, embedding, search
L1 foundation/ parsers, services, utils
L0 foundation/types/ constants, models, type definitions
Environment Variables
LLM (first match wins)
| Variable | Purpose | Default |
|---|---|---|
LLM_API_KEY |
Generic LLM key (highest priority) | - |
LLM_BASE_URL |
API endpoint | https://api.openai.com/v1 |
LLM_MODEL |
Model name | gpt-4o |
OPENAI_API_KEY |
OpenAI or compatible | - |
MOONSHOT_API_KEY |
Moonshot / Kimi (legacy) | - |
Embedding
| Variable | Purpose | Default |
|---|---|---|
DASHSCOPE_API_KEY |
DashScope (Qwen3 Embedding) | - |
DASHSCOPE_BASE_URL |
DashScope endpoint | https://dashscope.aliyuncs.com/api/v1 |
System
| Variable | Purpose | Default |
|---|---|---|
TERRAIN_WORKSPACE |
Workspace directory | ~/.terrain |
Installation Options
Install from PyPI
# Core (includes C/C++, Python, JS/TS grammars)
pip install terrain-ai
# With all language grammars (Rust, Go, Java, Scala, Lua)
pip install "terrain-ai[treesitter-full]"
Install from npm
# Global install (recommended for CLI usage)
npm install -g terrain-ai@latest
# Or run directly with npx (no install needed)
npx terrain-ai@latest --version
Install from local source
git clone https://github.com/JeremyJiao01/CodeGraphWiki.git
cd CodeGraphWiki
# Install with all language grammars
pip install ".[treesitter-full]"
# Or install in editable mode for development
pip install -e ".[treesitter-full]"
Build and install from wheel
git clone https://github.com/JeremyJiao01/CodeGraphWiki.git
cd CodeGraphWiki
python3 -m build
pip install dist/terrain_ai-*.whl
Development
git clone https://github.com/JeremyJiao01/CodeGraphWiki.git
cd CodeGraphWiki
pip install -e ".[treesitter-full]"
python3 -m pytest tests/ -v
# Integration tests (requires tinycc repo at ../tinycc)
python3 -m pytest tests/domains/core/test_graph_build.py -v # ~3 min
python3 -m pytest tests/domains/upper/test_api_docs.py -v # ~3 min
python3 -m pytest tests/domains/core/test_step3_embedding.py -v # ~27 min (API calls)
python3 -m pytest tests/domains/upper/test_api_find_integration.py -v # ~47 min (full pipeline)
License
Apache License 2.0 — see LICENSE for details.
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.