12 Best Graft Alternatives in 2026 (Free & Paid)
Graft is local codebase knowledge graph + MCP server that gives coding agents persistent structural context, priced at open source. If it is too expensive, missing a feature or simply not for you, these are the strongest alternatives we have reviewed.
1
GitHub's official MCP server: issues, pull requests, code, Actions and security alerts
★ 33k · +132 this week · MIT · updated 2026-10-02
9.0/10
What it does GitHub MCP Server is GitHub's own bridge between coding agents and the GitHub platform. Once connected, an agent can read repositories and files, search code, open and update issues and pull requests, look at GitHub Actions runs to work out why a build failed, and review code scanning, secret scanning and Dependabot findings. It is written in… Read more →
Pros
- Official and complete
- Remote or local
- Toolsets keep context small
Cons
- Broad permissions: scope your token carefully
Pricing: Free · github official
2
Build, run, test and debug iOS and macOS apps from AI agents, maintained by Sentry
★ 6.5k · +27 this week · MIT · updated 2026-10-02
8.4/10
What it does XcodeBuildMCP lets coding agents build, run, test and debug iOS and macOS projects. It is now maintained by Sentry. It wraps xcodebuild, the simulator tools, LLDB and Swift Package Manager in structured tools, so the agent can fix its own build errors, launch the app and check the result without you copying logs by hand. The same… Read more →
Pros
- Covers simulator, device, macOS, SwiftPM, LLDB and UI automation
- Workflow groups keep the tool list lean; also ships a CLI and Agent Skills
- Ready-made configs for most major agents
Cons
- macOS with Xcode 16+ only
- Error telemetry to Sentry is on by default (opt-out)
Pricing: Open source · ios macos xcode swift simulator
3
Sentry's MCP server: pull errors, stack traces and issues into your agent
★ 898 · +37 this week · FSL-1.1-Apache-2.0 · updated 2026-10-02
8.2/10
What it does Sentry MCP is Sentry's official server for connecting coding agents to your error and performance data. It is built for developers working alongside an agent, not as a general interface to every part of Sentry. A typical use: you paste an issue link or ask about the latest crash, and the agent pulls the stack trace, breadcrumbs… Read more →
Pros
- Real error context for fixes
- Official, hosted with OAuth
Pricing: Freemium · errors monitoring
🔥 Deal: Sentry for Startups: up to $5,000 in credits (12-month expiry) for companies founded in the last 2 years with under $5M raised
4
The official reference MCP servers: fetch, filesystem, git, memory and more
★ 91k · +362 this week · MIT / Apache-2.0 · updated 2026-10-01
8.0/10
What it does The modelcontextprotocol/servers repository contains the reference servers maintained by the MCP steering group. They exist mainly to show how the protocol and the official SDKs work, and several are also handy in everyday use. The repository used to hold many more servers, for services such as GitHub, Slack, PostgreSQL, Puppeteer and Brave Search. Those have been moved… Read more →
Pros
- Canonical examples of MCP
- Small, simple servers
Cons
- Reference quality, not all production-hardened
Pricing: Free · official reference
5
Lets AI agents drive the Unity Editor: scenes, C# scripts, assets, tests and builds
★ 15k · +173 this week · MIT · updated 2026-10-02
8.0/10
What it does MCP for Unity connects an AI coding agent to a running Unity Editor. It has two parts: a C package that you add to your Unity project, and a Python server that your MCP client talks to. With both running, the agent can create and change scenes and GameObjects, write and edit C scripts, manage assets, prefabs… Read more →
Pros
- Broad tool set covering scenes, scripts, assets, builds and profiling
- In-Editor setup window auto-configures detected MCP clients
- Tool groups keep the default tool list small
Cons
- Needs both a Unity package and a Python/uv server running
- AI asset generation tools depend on paid third-party keys
Pricing: Open source · unity game-development csharp editor-automation 3d
6
n8n node docs, templates and validation so agents build workflows that actually run
★ 23k · +31 this week · MIT · updated 2026-10-01
7.8/10
What it does n8n-MCP helps a coding agent build n8n automation workflows that actually run. On its own, an agent tends to guess node names and parameters and produce workflow JSON that breaks at runtime. This server gives it a searchable local database of n8n's core and community nodes, including property schemas, operations, documentation and real configuration examples, plus a… Read more →
Pros
- Schema-level node documentation plus workflow and expression validation
- Optional instance tools create, test and fix workflows
- Tools can be disabled one by one for governance
Cons
- Telemetry is on by default and may be used to train models
- With an API key the agent can delete workflows and manage credentials
Pricing: Freemium · n8n workflow-automation validation templates low-code
🔥 Deal: Startup Plan: 50% off the self-hosted Business plan (fewer than 20 employees, under €5M raised)
7
Official MathWorks server to run, test and lint MATLAB code from AI agents (needs a MATLAB licence)
★ 1.6k · +30 this week · updated 2026-09-25
7.8/10
What it does The MATLAB MCP Server is MathWorks' official way to let AI coding agents use a local MATLAB installation. The agent can start or attach to MATLAB, run code snippets or whole .m scripts, run unit tests and check code with MATLAB's static analyser. It gets the real output back, so it can write MATLAB, execute it, read… Read more →
Pros
- Official MathWorks server with code run, test and static-analysis tools
- Can attach to your open MATLAB session
- Ships MATLAB coding guidelines as MCP resources
Cons
- Requires a licensed local MATLAB install (R2021a or later); the server licence restricts use to MathWorks products and one user
- Agent can run arbitrary MATLAB code; usage telemetry on by default
Pricing: Free · matlab scientific-computing testing static-analysis mathworks
8
MCP server giving an AI agent terminal control, filesystem search and diff-based file editing
★ 9.9k · +130 this week · MIT · updated 2026-10-01
7.2/10
What it does Desktop Commander is an MCP server that gives an AI assistant hands-on control of your local machine: a terminal, a filesystem, and surgical code editing, rather than just a chat window. It was built with Claude Desktop in mind but speaks plain MCP, so any MCP-capable client can use it. Tools it gives your agent - Terminal… Read more →
Pros
- Broad client support: Claude Desktop/web, ChatGPT, Cursor, Windsurf, VS Code/Copilot, Cline and JetBrains
- Diff-based editing avoids clumsy full-file rewrites
- Active, popular project with frequent commits
Cons
- Grants real terminal and filesystem access, so it needs careful scoping on a trusted machine
- Primary docs and examples are Claude-Desktop-centric even though other clients are supported
Pricing: Open source · mcp terminal filesystem code-editing claude-desktop
9
jscpd Open source
Fast duplicate-code detector with an MCP server so agents can check for copy-paste before committing
★ 6.3k · +51 this week · MIT · updated 2026-10-02
7.0/10
What it does jscpd is a mature, Rust-based duplicate-code detector that tokenizes source files across 224+ languages/formats and uses a Rabin-Karp rolling hash to find copy-pasted (Type-2/Type-3) code blocks. It also flags dead code, ranks files by complexity, tracks duplication trends through git history, and rolls results into a composite codebase health score. Tools it gives your agent - An… Read more →
Pros
- Mature, battle-tested tool (MIT, 6.2k stars, 10M+ monthly npm downloads) already depended on by ~5,000 repos including projects from Google, Microsoft and Sales
- Genuine MCP server (stdio) plus an installable agent skill let a coding agent check a snippet against the codebase before adding more duplication
- Fast Rust engine (224+ languages/formats), SARIF/GitHub Code Scanning support, and a compact AI-oriented report format
Cons
- Primarily a long-standing traditional linter/duplication scanner — the AI-native surface (MCP server, agent skill) is a secondary feature layered on afterward,
- Default install path is a curl-pipe-to-shell script (from the project's own domain, jscpd.dev), which some teams may want to avoid in favor of the npm/Homebrew
Pricing: Open source · duplicate-code static-analysis mcp-server ci-cd code-quality
10
AST-based local semantic code search CLI and MCP server that cuts agent context tokens
★ 2.7k · +11 this week · Apache-2.0 · updated 2026-10-01
7.0/10
What it does CocoIndex Code (ccc) is a lightweight, Rust-powered semantic code search engine. Instead of grepping for exact strings, it builds an AST-based index of your repository so you — or your AI coding agent — can search it using natural-language descriptions of what you're looking for. Tools it gives your agent It gives an agent a fast, local… Read more →
Pros
- Real engineering behind it: Rust-based AST indexing across 28+ languages, works fully offline with the [full] local-embeddings variant, Apache-2.0 licensed
- Broad, concretely documented agent integration: Claude Code (skill or MCP), Cursor, Codex, OpenCode, Kilo Code, and a Grok plugin with hooks
- Claims ~70% token reduction for agent code search versus naive text search, with source code never leaving the machine
Cons
- Younger/smaller project (~2.7k stars, 247 commits) than established alternatives
- Enables anonymous usage telemetry by default (opt-out via COCOINDEX_DISABLE_USAGE_TRACKING=1)
Pricing: Open source · semantic-search mcp-server cli ast code-search
11
Control and observe real ROS/ROS2 robots from Claude, Cursor or Codex
★ 1.5k · +3 this week · Apache-2.0 · updated 2026-10-02
7.0/10
What it does ROS MCP Server is an open-source Model Context Protocol bridge between AI coding agents and robots running ROS (Robot Operating System) or ROS2. It exposes your robot's live topics, services, actions and parameters to an MCP client so an LLM can inspect and control a real or simulated robot in the same session it uses to write… Read more →
Pros
- Bidirectional bridge: agents can both command and observe live robot state, not just fire-and-forget commands
- No changes needed to existing robot code - just add the rosbridge node
- Actively maintained with 400+ commits, real issue and PR activity, and Apache-2.0 licensing
Cons
- Robotics-specific: only useful if you actually have a ROS/ROS2 stack, not a general-purpose coding tool
- Requires a working rosbridge setup alongside the MCP server, adding a moving part to configure
Pricing: Free · mcp robotics ros ros2 hardware
12
AI-driven reverse engineering: 250+ MCP tools over Ghidra's decompiler, analysis and debugger
★ 4.1k · +105 this week · Apache-2.0 · updated 2026-10-02
6.5/10
What it does Ghidra MCP Server bridges Ghidra's reverse-engineering toolkit with AI coding agents via the Model Context Protocol. It ships as a Ghidra GUI plugin plus a Python bridge/headless server, giving an agent structured, high-level access to binary analysis, decompilation and documentation workflows instead of requiring a human to script Ghidra's Java API directly. Tools it gives your agent… Read more →
Pros
- Enormous, well-organized tool surface (~250 tools) spanning decompilation, data types, cross-refs, malware/IOC analysis, P-code emulation and live TraceRmi debu
- Actively developed: v7.0.0, 1,190+ commits, 3.9k stars, 11 open PRs — not an abandoned side project
- Batch operations and lazy per-group tool loading keep large tool counts usable by real MCP clients
Cons
- Requires a local, version-matched Ghidra install (12.1.2) plus Java 21 and a Maven/uv build toolchain — a much heavier setup than a typical hosted MCP server
- Given its reach into script execution and live debugging, remote/networked use needs explicit auth and script execution is disabled by default — worth understan
Pricing: Open source · reverse-engineering ghidra binary-analysis decompiler malware-analysis