context-squeeze MCP server
Commit Nov 11, 2026Provides AI agents with a compact symbol map of a codebase to navigate files without consuming excessive context window space.
Also known as Ankara-Software/context-squeeze
- Stars
- 1
- Tools
- 4
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 11, 2026
- License
- MIT
What is context-squeeze MCP?
context-squeeze provides AI agents with a compact symbol map of a codebase to navigate files without consuming excessive context window space. It exposes 4 tools, such as path, file_path, symbol_name.
Summary generated from the project's README
About context-squeeze MCP server
Navigating large codebases without exhausting your AI model's context window is the primary function of context-squeeze. This Rust-based MCP server allows agents like Claude Desktop, Cursor, or Cline to explore repositories by first requesting a compact symbol map containing signatures for classes, structs, interfaces, and functions. Once the agent identifies the relevant logic, it can fetch the full body of a specific symbol rather than reading entire files.
Developers use this tool by installing it from source via the Rust toolchain and configuring it as a stdio server in their client settings. It supports Python, TypeScript, JavaScript, Rust, and Go files. The server provides three specific tools: get_repo_map for generating a workspace outline, get_code_item for retrieving specific symbol bodies, and search_code for performing case-insensitive substring searches across symbol names and signatures. Because it uses regular expressions for symbol extraction, it is lightweight but may miss certain internal methods in specific languages.
What you can do with it
- 1Generate a workspace outlineAsk the AI to create a compact map of all classes and functions in a directory to understand the project structure.
- 2Retrieve specific function logicRequest the full implementation of a single function or struct once its signature has been identified.
- 3Search for specific symbolsFind where a particular interface or function is defined by searching for a substring across the entire codebase.
Written from the project's README. Every use case is checked against a quote from it.
context-squeeze MCP tools 4
Detected from the READMEpathfile_pathsymbol_namequery
| path | — | |
| file_path | — | |
| symbol_name | — | |
| query | — |
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachableankara-software/context-squeeze is public and reachablegithub.com/ankara-software/context-squeeze ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-11github.com/ankara-software/context-squeeze ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model4 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubMaintained by Ankara Yazılım | Project page
Status: beta. It builds, the 5 parser unit tests pass and the three MCP tools were called over stdio against sample Python, TypeScript and Go files. Symbol extraction uses regular expressions, not a real parser, so it misses some symbols (see Limitations). No prebuilt binaries or npm package have been published yet; install from source.
Context-Squeeze
Context-Squeeze is a small Model Context Protocol (MCP) server written in Rust that helps AI agents (such as Claude Desktop, Cursor, Cline or local LLM clients) navigate a codebase. Instead of reading whole files, the agent first asks for a compact symbol map (signatures of classes, structs, interfaces and functions) and then fetches the body of only the symbol it needs. This keeps less source text in the model's context window.
Türkçe
Context-Squeeze, yapay zeka ajanlarının (Claude Desktop, Cursor, Cline vb.) büyük kod tabanlarında gezinmesine yardım eden, Rust ile yazılmış küçük bir Model Context Protocol (MCP) sunucusudur. Ajan, dosyaların tamamını okumak yerine önce sınıf, struct, arayüz ve fonksiyon imzalarından oluşan kısa bir sembol haritası ister, sonra yalnızca ihtiyaç duyduğu sembolün gövdesini getirir. Böylece modelin bağlam penceresinde daha az kaynak kodu tutulur. Python, TypeScript/JavaScript, Rust ve Go dosyalarını destekler; sembolleri düzenli ifadelerle çıkarır, bu yüzden bazı semboller (örneğin Python ve TypeScript sınıf metotları) haritada görünmez. Proje beta aşamasındadır; hazır ikili dosya ve npm paketi henüz yayımlanmamıştır, kaynaktan kurulur.
Installation
Build from source (Rust toolchain required):
cargo install --git https://github.com/Ankara-Software/context-squeeze
This installs a context-squeeze binary that speaks MCP over stdin and stdout. Or build it manually:
git clone https://github.com/Ankara-Software/context-squeeze.git
cd context-squeeze
cargo build --release
# The compiled binary will be in target/release/context-squeeze
The package.json and bin/cli.js files describe an npm wrapper (context-squeeze-mcp) that downloads a release binary. That package is not on the npm registry and the v0.1.0 release has no binary assets, so npx -y context-squeeze-mcp does not work yet.
Client Setup
Claude Desktop
Add the following to your claude_desktop_config.json (%APPDATA%\Claude\claude_desktop_config.json on Windows, ~/Library/Application Support/Claude/claude_desktop_config.json on macOS). Use the full path of the binary if it is not on your PATH:
{
"mcpServers": {
"context-squeeze": {
"command": "context-squeeze"
}
}
}
Cursor
- Open Cursor Settings -> Features -> MCP.
- Click + Add New MCP Server.
- Set:
- Name:
context-squeeze - Type:
stdio - Command:
context-squeeze
- Name:
- Save.
get_code_item and search_code read files relative to the working directory of the server process, so start the server from the project you want to explore.
Exposed MCP Tools
1. get_repo_map
Walks the workspace and builds a compact outline. It respects .gitignore rules and skips hidden files.
- Returns: for each file with symbols, the file path and the signature line of each class, struct, interface, trait, enum and function found. Bodies are omitted.
- Arguments:
path: (Optional) Directory to scan. Defaults to the current directory.
2. get_code_item
Returns the full body of one symbol in one file.
- Arguments:
file_path: (Required) Path to the source file, relative to the working directory.symbol_name: (Required) Name of the class, struct, function or interface.
3. search_code
Case-insensitive substring search over symbol names and signatures across the workspace.
- Arguments:
query: (Required) Text to look for.
- Returns: file, symbol, kind, signature and line range for each match.
Supported languages
Files with these extensions are indexed: py, ts, tsx, js, jsx, rs, go.
Limitations
- Symbols are found with regular expressions. Methods inside Python and TypeScript classes are not listed (the class line is); Rust
implfunctions and Go methods are. - Only the extensions listed above are indexed.
- The server does not restrict paths:
get_code_itemjoins the given path to the working directory. Run it only for clients and projects you trust.
Measured Example
One run of get_repo_map on this repository's own source plus three small sample files (28,853 bytes of source in total) returned a map of 1,968 bytes. This is a byte count from a single run, not a token count, and the ratio depends heavily on the codebase.
Development
cargo build
cargo test
See CONTRIBUTING.md, SECURITY.md and CODE_OF_CONDUCT.md.
License
MIT. Copyright (c) 2026 Anılcan Kara and Ankara Yazılım contributors.
context-squeeze MCP questions
What tools does context-squeeze MCP provide?
It exposes 4 tools, including path, file_path, symbol_name, query.
What programming languages are supported?
The server indexes files with py, ts, tsx, js, jsx, rs, and go extensions.
How does the symbol extraction work and what are its limitations?
It uses regular expressions rather than a real parser, meaning it may miss certain symbols like methods inside Python and TypeScript classes.