Context Engine MCP Server

A high-performance, memory-safe Model Context Protocol (MCP) server written in Rust for AI-powered codebase context retrieval.

Overview

Context Engine provides semantic code search and AI-powered context retrieval for coding agents. It integrates with the Augment Code SDK to offer:

• Semantic Code Search: AI-powered codebase retrieval using embeddings
• File Indexing: Automatic workspace indexing with intelligent file filtering
• MCP Protocol: Full Model Context Protocol support (JSON-RPC over stdio/HTTP)
• Code Review: Multi-pass review pipeline with risk scoring and invariant checking
• Planning: AI-assisted task planning and step management
• Memory: Persistent memory storage for context across sessions

Features

Metric	Value
Binary Size	~7 MB (optimized ARM64)
Lines of Code	~10,500 Rust
Unit Tests	201 tests
Integration Tests	11 tests
MCP Tools	72 tools
Agent Skills	7 skills
Supported Languages	18+ (symbol detection)
Startup Time	<10ms
Memory Usage	~20 MB idle

Installation

Prerequisites

• Rust 1.83+ (with cargo)
• Augment API credentials (via ~/.augment/session.json or environment variables)

Build from Source

cargo build --release

The binary will be at target/release/context-engine.

Usage

Command Line

# Start MCP server (stdio transport - default)
./target/release/context-engine --workspace /path/to/project
# Start with HTTP transport
./target/release/context-engine --workspace /path/to/project --transport http --port 3000
# Enable metrics endpoint
./target/release/context-engine --workspace /path/to/project --metrics --metrics-port 9090
# Debug mode
./target/release/context-engine --workspace /path/to/project --debug

Environment Variables

Variable	Description
AUGMENT_API_TOKEN	API authentication token
AUGMENT_API_URL	API base URL
CONTEXT_ENGINE_DEBUG	Enable debug logging

Configuration

Credentials are resolved in order:

1. Command-line options
2. Environment variables
3. Session file (~/.augment/session.json)

MCP Tools (72 Total)

Retrieval Tools (7)

Tool	Description
codebase_retrieval	Semantic search across the codebase
semantic_search	Search for code patterns and text
get_file	Retrieve file contents with optional line range
get_context_for_prompt	Get comprehensive context bundle
enhance_prompt	AI-powered prompt enhancement with context injection
bundle_prompt	Bundle raw prompt with codebase context (no AI rewriting)
tool_manifest	Discover available capabilities

Index Tools (5)

Tool	Description
index_workspace	Index files for semantic search
index_status	Check indexing status
reindex_workspace	Clear and rebuild index
clear_index	Remove index state
refresh_index	Refresh the codebase index

Memory Tools (6)

Tool	Description
store_memory	Store persistent memories
retrieve_memory	Recall stored memories
list_memory	List all memories
delete_memory	Delete a memory
memory_store	Store with rich metadata (kind, language, tags, priority)
memory_find	Hybrid search with filtering

Planning Tools (20)

Tool	Description
create_plan	Create AI-powered implementation plans
get_plan	Get plan details
list_plans	List all plans
add_step	Add a step to a plan
update_step	Update step status
refine_plan	Refine plan with AI
visualize_plan	Generate visual representation
execute_plan	Execute plan steps
save_plan	Save plan to storage
load_plan	Load plan from storage
delete_plan	Delete a plan
start_step	Mark step as in progress
complete_step	Mark step as completed
fail_step	Mark step as failed
view_progress	View plan progress
view_history	View execution history
request_approval	Create approval request
respond_approval	Respond to approval request
compare_plan_versions	Generate diff between versions
rollback_plan	Rollback to previous version

Review Tools (14)

Tool	Description
review_diff	Review code changes with risk analysis
analyze_risk	Analyze risk level of changes
review_changes	Review code changes in files
review_git_diff	Review current git diff
review_auto	Automatically review recent changes
check_invariants	Check code invariants
run_static_analysis	Run static analysis
scrub_secrets	Scan for potential secrets
validate_content	Validate content against rules
get_review_status	Get review status
reactive_review_pr	Start parallelized PR review
pause_review	Pause a running review session
resume_review	Resume a paused review session
get_review_telemetry	Get detailed review metrics

Navigation Tools (3)

Tool	Description
find_references	Find all references to a symbol
go_to_definition	Navigate to symbol definition
diff_files	Compare two files with unified diff

Workspace Tools (7)

Tool	Description
workspace_stats	Get workspace statistics and metrics
git_status	Get current git status
extract_symbols	Extract symbols from a file
git_blame	Get git blame information
git_log	Get git commit history
dependency_graph	Generate dependency graph
file_outline	Get file structure outline

Specialized Search Tools (7)

Tool	Description
search_tests_for	Find test files with preset patterns
search_config_for	Find config files (yaml/json/toml/ini/env)
search_callers_for	Find callers/usages of a symbol
search_importers_for	Find files importing a module
info_request	Simplified retrieval with explanation mode
pattern_search	Structural code pattern matching
context_search	Context-aware semantic search

Skills Tools (3)

Tool	Description
list_skills	List all available Agent Skills
search_skills	Search skills by query (metadata only)
load_skill	Load full skill instructions on demand

Agent Skills

Context Engine implements the Tool Search Tool pattern for progressive disclosure of Agent Skills. This reduces token overhead by ~75% compared to loading all tool definitions upfront.

Available Skills

Skill	Category	Description
planning	workflow	Task planning and execution for complex multi-step tasks
code_review	quality	Comprehensive code review workflow
search_patterns	search	Specialized search patterns for tests, configs, callers
debugging	troubleshooting	Systematic debugging workflow for identifying and fixing bugs
refactoring	quality	Safe code refactoring workflow with impact analysis
documentation	quality	Documentation generation and maintenance workflow
testing	quality	Comprehensive test writing and maintenance workflow

How Skills Work

1. Discovery: Call list_skills() or search_skills(query) to find relevant skills
2. Loading: Call load_skill(id) to get full instructions
3. Execution: Follow the skill instructions using primitive MCP tools
4. Via Prompts: Skills are also available as MCP prompts (e.g., skill:debugging)

Skills are loaded from skills/ directory as SKILL.md files following the Agent Skills specification.

Client Compatibility

Client	How Skills Are Accessed
Claude Code	Native Agent Skills support (reads SKILL.md directly)
Cursor	MCP tools (search_skills, load_skill)
GitHub Copilot	AGENTS.md + MCP tools
Windsurf	MCP tools
VS Code + Continue	MCP prompts (skill:*)
OpenAI Codex	AGENTS.md

Architecture

src/
├── main.rs # Entry point with CLI
├── lib.rs # Library exports
├── error.rs # Error types
├── config/ # Configuration management
├── sdk/ # Augment API client
│ ├── api_client.rs # HTTP client
│ ├── blob.rs # SHA256 blob naming
│ ├── credentials.rs # Auth resolution
│ └── direct_context.rs # Context operations
├── service/ # Business logic layer
│ ├── context.rs # Context service
│ ├── memory.rs # Memory service
│ └── planning.rs # Planning service
├── mcp/ # MCP protocol layer
│ ├── server.rs # MCP server
│ ├── handler.rs # Request handler
│ ├── protocol.rs # JSON-RPC types
│ └── transport.rs # Stdio/HTTP transports
├── tools/ # MCP tool implementations
├── reviewer/ # Code review pipeline
├── reactive/ # Reactive review system
├── watcher/ # File system watcher
├── http/ # HTTP server (axum)
├── metrics/ # Prometheus metrics
└── types/ # Shared type definitions

Docker Support

The Docker image is ~20 MB (Alpine-based with statically linked binary).

Build Docker Image

docker build -t context-engine .

Run with Docker

# HTTP mode
docker run -d \
-v /path/to/project:/workspace:ro \
-v ~/.augment:/home/context-engine/.augment:ro \
-p 3000:3000 \
-p 9090:9090 \
context-engine \
--workspace /workspace \
--transport http \
--metrics
# Stdio mode (for MCP integration)
docker run -i \
-v /path/to/project:/workspace:ro \
-v ~/.augment:/home/context-engine/.augment:ro \
context-engine \
--workspace /workspace

Docker Compose

# Set your project pathexport PROJECT_PATH=/path/to/project
# Start services
docker-compose up -d
# View logs
docker-compose logs -f
# Stop
docker-compose down

Development

Running Tests

# Run all unit tests (170 tests)
cargo test --lib
# Run integration tests (basic CLI tests)
cargo test --test mcp_integration_test
# Run full integration tests including MCP protocol tests
cargo test --test mcp_integration_test -- --ignored
# Run all tests
cargo test --all-targets

Test Categories

Category	Count	Description
Unit Tests	201	Core functionality tests
Integration Tests	11	MCP protocol and CLI tests

Linting

cargo clippy --all-targets --all-features -- -D warnings

Formatting

cargo fmt

Code Coverage

# Install cargo-tarpaulin
cargo install cargo-tarpaulin
# Run with coverage
cargo tarpaulin --out Html

MCP Client Configuration

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
"mcpServers": {
"context-engine": {
"command": "/path/to/context-engine",
"args": ["--workspace", "/path/to/your/project"]
}
}
}

Cursor

Add to your MCP configuration:

{
"context-engine": {
"command": "/path/to/context-engine",
"args": ["--workspace", "."]
}
}

License

MIT License - See LICENSE file for details.