What It Is

CostAffective is a local MCP server that helps AI coding assistants understand large codebases without sending your code anywhere.

It provides:

• repository-aware retrieval
• symbol and reference lookup
• caller discovery
• repository summaries
• fast re-indexing
• client installation and diagnostics

No API key is required.

Installation

macOS / Linux / Windows with Go installed

go install github.com/okyashgajjar/costaffective-mcp@latest

This works on all three platforms. After installation, open a new terminal if costaffective is not yet on your PATH.

Build from source

# macOS / Linux
go build -o costaffective ./cmd/mycli/

# Windows
go build -o costaffective.exe ./cmd/mycli/

If you build on Windows, run the binary as .\costaffective.exe.

Connect your clients

costaffective install --all

On Windows, use:

.\costaffective.exe install --all

The installer detects supported clients and writes the correct MCP config for each one.

macOS / Linux shell installer

bash install-mcp.sh

Supported Platforms

• Windows
• macOS
• Linux

Supported Clients

| Client | Config Surface | |--------|----------------| | Claude Code | ~/.claude.json, .mcp.json, or settings files | | Cursor | ~/.cursor/mcp.json or workspace MCP settings | | OpenCode | opencode.json | | Codex CLI | ~/.codex/config.toml | | Antigravity | ~/.gemini/config/mcp_config.json | | MCP-compatible clients | stdio transport |

MCP Tools

| Tool | What it does | |------|--------------| | search_code | Semantic code search backed by tree-sitter parsing | | find_symbol | Find where a symbol is defined | | find_references | Find every use of a symbol | | find_callers | Find functions that call a target function | | grep_code | Regex and text search fallback | | get_repository_summary | Summarize modules, files, languages, and architecture | | index_repository | Rebuild or refresh the repository index |

Commands

| Command | Description | |---------|-------------| | costaffective install | Interactive installation | | costaffective install --all | Configure every detected client | | costaffective install --target <name> | Configure one client only | | costaffective install --repair | Repair the binary and MCP configuration | | costaffective doctor | Validate installation and startup | | costaffective uninstall | Remove MCP configs from clients | | costaffective serve | Start the MCP stdio server |

Quick Start

# Install the binary
go install github.com/okyashgajjar/costaffective-mcp@latest

# Connect the supported clients
costaffective install --all

# Check the install
costaffective doctor

Doctor

costaffective doctor checks:

• binary existence and permissions
• PATH visibility
• MCP configuration for each client
• server startup
• repository state

Repository State

CostAffective keeps track of the repository index and the working tree:

• unindexed means no usable index exists yet
• stale means files changed after indexing
• ready means the repository is aligned with the index

Agent mode can auto-index when needed; other modes can prompt first.

Architecture

AI Client (MCP Host)
    │
    ├── stdio transport ──► costaffective serve (MCP Server)
    │                           │
    │                           ├── search_code ───────────► tree-sitter AST match
    │                           ├── find_symbol ───────────► SymbolDB lookup
    │                           ├── find_references ───────► SymbolDB reference search
    │                           ├── find_callers ──────────► SymbolDB call graph
    │                           ├── grep_code ─────────────► full-text search fallback
    │                           ├── get_repository_summary ► KnowledgeStore
    │                           └── index_repository ──────► SharedIndexer

Why CostAffective

Modern coding agents waste context by repeatedly rediscovering the same code paths.

CostAffective keeps the repository index local and gives the model smaller, more relevant context so it can spend tokens on reasoning instead of discovery.

Benchmark Highlights

Continue OSS Repository

| Metric | Value | |--------|-------| | Files | 3,203 | | Source Files | 1,985 |

CostAffective

| Metric | Value | |--------|-------| | Tokens | 4.7M | | API Calls | 89 | | Exploration Calls | 43 |

Alternative Semantic Code Intelligence Benchmark

| Metric | Value | |--------|-------| | Tokens | 8.7M | | API Calls | 134 | | Exploration Calls | 94 |

Observed Results

| Metric | Improvement | |--------|-------------| | Token Usage | 45.9% lower | | Exploration Loops | 54.3% lower | | Tool Interactions | 42.1% lower |

Development

git clone https://github.com/okyashgajjar/costaffective-mcp.git
cd costaffective-mcp
go build ./...
go test ./...

License

MIT