Gortex – MCP server for cross-repo code intelligence

TL;DR

TL;DR: Gortex is an open-source code-intelligence MCP server that indexes multiple repositories into a graph, giving AI coding agents symbol lookup, call chains, blast-radius analysis, and cross-repo contract detection with up to 50× fewer tokens per response.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: get.gortex.dev
• Source repository: github.com/zzet/gortex
• License: Apache-2.0 (verified via GitHub API license.spdx_id)
• HN launch thread: news.ycombinator.com/item?id=47795300 — Show HN
• Source last checked: 2026-06-26 (commit main, README wall.svg hero confirmed present)

What Is Gortex?

Gortex is a high-performance code-intelligence engine that parses repositories into a persistent, provenance-tiered knowledge graph and exposes it via an MCP server, CLI, and web UI.

Rather than having an AI agent blast-read every file in a repo (wasting context tokens), Gortex lets agents ask precise graph queries — “what calls this function?”, “what breaks if I change this API?”, “show me the blast radius of this refactor” — and get back just the relevant nodes.

The project’s README makes a strong claim: up to 50× fewer tokens per response compared to naive file reads. This comes from graph-native lookups that return only the symbols and edges an agent actually needs, not surrounding noise.

Key claims from the README:

• 175 MCP tools across symbol lookup, call chains, blast radius, dataflow, clone detection, refactoring, and code actions
• 257 languages/grammars via tree-sitter AST analysis
• 17 AI coding agents supported out of the box: Claude Code, Kiro, Cursor, Windsurf, VS Code/Copilot, Continue.dev, Cline, OpenCode, Antigravity, Codex CLI, Gemini CLI, Zed, Aider, Kilo Code, OpenClaw, Hermes, and Oh My Pi
• Cross-repo by default — N repos in one graph; contracts, references, and call chains span repo boundaries
• Single static binary — no dependency chain for macOS, Linux, and Windows
• Zero external dependencies — everything in-process, no network required to get started
• GCX1 wire format — a published, round-trippable binary format that cuts an additional 27% tokens vs JSON at the same fidelity

Setup Workflow

Step 1: Install Gortex

# macOS / Linux
curl -fsSL https://get.gortex.dev | sh

# Windows (PowerShell)
irm https://get.gortex.dev/install.ps1 | iex

The install script detects your OS and architecture, verifies SHA256 and cosign signatures, and places the binary on your PATH. Re-run to upgrade.

Additional install methods available: Homebrew, .deb/.rpm/.apk packages, scoop, and signed binaries. From-source builds are also documented in the repo.

Step 2: Initialize machine and start daemon

gortex install                          # one-time machine setup (MCP, skills, slash commands)
gortex daemon start --detach           # background daemon

Step 3: Track and index repositories

gortex track ~/projects/myapp           # add a repo to the graph
cd ~/projects/myapp && gortex init      # per-repo: .mcp.json, hooks, community routing

Step 4: Verify integration with your AI agent

After gortex init, the tool auto-detects which coding agents are installed on your machine and configures them to use graph queries. Check the agents list:

gortex agents list

Supported agents (from docs/agents.md): Claude Code, Kiro, Cursor, Windsurf, VS Code/Copilot, Continue.dev, Cline, OpenCode, Antigravity, Codex CLI, Gemini CLI, Zed, Aider, Kilo Code, OpenClaw, Hermes, Oh My Pi.

Step 5: Query via MCP or CLI

# CLI example: find all callers of a function
gortex call-chains --symbol myFunction --depth 3

# MCP tool (for AI agents): blast_radius, contracts, clone_detection, etc.

Deeper Analysis

Graph-native vs. naive file reads

The token-efficiency claim deserves scrutiny. Naive RAG or context-injection strategies dump entire files into context — even when the agent only needs to understand a single function signature or call site. Gortex replaces this with graph traversals that return only the relevant nodes and edges.

The 50× figure is backed by reproducible benchmarks in BENCHMARK.md in the repo.

Cross-repo contract detection

Gortex auto-detects API contracts across repository boundaries:

| Contract type | Detection method | Example | |---|---|---| | HTTP routes | Framework annotations (gin, Express, FastAPI, Spring…) | Route handler vs. HTTP client calls | | gRPC | Proto service definitions | Service RPC vs. client stub calls | | GraphQL | Schema type/field definitions | Schema vs. query strings | | Message topics | Kafka/RabbitMQ/NATS/Redis pub/sub | Publish calls vs. subscribe calls | | WebSocket | Event emit/listen patterns | emit() vs. on() | | Env vars | os.Getenv, process.env, .env files | Setenv vs. Getenv |

Languages and grammar coverage

Gortex supports 257 languages via tree-sitter AST analysis, augmented with compiler-grade resolution for Python, TypeScript/JavaScript, PHP, C#, Go, C, C++, Java, Kotlin, Swift, Zig, Rust, Ruby, Elixir, OCaml, and Haskell. A three-tier approach (bespoke tree-sitter, regex, forest-backed signatures) plus Jupyter and Databricks notebook support ensures broad coverage.

Security and telemetry

Telemetry is off by default. When enabled, only anonymized tool/command counts are transmitted — no code, paths, names, or exact counts. The DO_NOT_TRACK environment variable is respected.

SLSA Level 3 supply-chain attestation, Sigstore signing for releases, and VirusTotal scanning of binaries are all documented in the repo.

Practical Evaluation Checklist

• [ ] Install completes without dependency errors
• [ ] gortex daemon start --detach starts successfully
• [ ] gortex track ~/your-repo indexes without errors
• [ ] gortex agents list shows your installed agent
• [ ] MCP tools respond correctly (test call-chains, blast-radius)
• [ ] Cross-repo tracking works when multiple repos are added
• [ ] Telemetry stays off by default (verify with gortex telemetry status)

Security Notes

• Telemetry off by default — verify before enabling
• SHA256 + cosign verification on install script
• SLSA Level 3 supply chain attestation for releases
• Single static binary — no runtime dependencies that could be tampered with
• Code never leaves your machine unless you explicitly configure an endpoint

FAQ

Q: Does it work with local-only repositories with no internet? A: Yes. Gortex is a single static binary with everything in-process. No network access is required to index or query repositories.

Q: Which AI coding agents are actually supported? A: The README lists 17: Claude Code, Kiro, Cursor, Windsurf, VS Code/Copilot, Continue.dev, Cline, OpenCode, Antigravity, Codex CLI, Gemini CLI, Zed, Aider, Kilo Code, OpenClaw, Hermes, and Oh My Pi. Run gortex agents list to see what was auto-detected on your machine.

Q: How does it compare to GitHub Copilot or Cursor’s built-in code intelligence? A: Gortex is an MCP server that supplements (not replaces) IDE-level code intelligence. It focuses on graph-native cross-repo queries — blast radius, call chains, contract detection — that general IDE tools don’t handle well. The 50× token claim specifically applies to how much context an agent uses per query compared to dumping full files.

Q: Is the 50× token reduction real? A: The README links to reproducible benchmarks in BENCHMARK.md. The claim applies to graph-native lookups vs. naive file reads — your mileage varies depending on query type.

Conclusion

Gortex solves the context-window bloat problem for AI coding agents by replacing naive file reads with precise graph queries. The MCP server interface means it works with any agent that supports MCP — and with 17 agents auto-detected out of the box, setup is minimal.

If you regularly work across multiple repositories, maintain large codebases, or find your agents burning tokens reading entire files just to answer a single question, Gortex is worth evaluating. The cross-repo contract detection alone is a unique capability that most alternatives don’t offer.

Install it in under a minute with the one-liner above, or read the full onboarding guide at docs/onboarding.md.