Spokes
Local MCP memory of your GitHub repos for Cursor
README
<div align="center">
<img src="assets/logo.svg" alt="Spokes logo" width="96" height="96" />
Spokes
Local MCP memory of your GitHub repositories for Cursor.
Your agent reuses code you already wrote instead of reinventing the wheel.
<br/>
<a href="#-quick-start"><img src="https://img.shields.io/badge/Quick%20Start-3DB8A8?style=for-the-badge&logoColor=white" alt="Quick Start" /></a> <a href="#-prerequisites"><img src="https://img.shields.io/badge/Prerequisites-152024?style=for-the-badge" alt="Prerequisites" /></a> <a href="#-mcp-tools"><img src="https://img.shields.io/badge/MCP%20Tools-152024?style=for-the-badge" alt="MCP Tools" /></a>
<br/>
<br/>
<img src="assets/banner.png" alt="Spokes — reuse your own code" width="100%" />
</div>
Why Spokes
AI agents love to reinvent things you already built. You have working auth, a music catalog, a UI kit, a bot — scattered across your GitHub repos — but the agent writes it all again from scratch, slightly differently, every time.
Spokes gives the agent a memory of your repositories. Before writing new code, it checks what you already have, pulls the useful spokes, and adapts them into the current project. If nothing fits, it codes as usual.
Think of each of your repos as a spoke. Spokes brings the right ones into the hub — your active project.
Features
- Repo memory cards — short summary, tags, tree overview, and entry points per repo
- On-demand code access — files & trees fetched from GitHub only when needed
- Reuse journal — every search and every piece taken is logged
- Live Reuse flow UI — real-time diagram of
task → repo → code → logic → result - Local-first — runs on your machine; token stays in local SQLite
- GitHub OAuth (Device Flow) — connect once, no Personal Access Token juggling
Prerequisites
You need these before (or during) install:
| Requirement | Required? | Notes |
|---|---|---|
| Windows 10/11 | Yes | Installer is PowerShell / .bat based |
| Cursor | Yes | MCP must be enabled |
| GitHub account | Yes | Used for Connect GitHub (Device Flow) |
| Git | Recommended | Faster installs; ZIP fallback exists |
| Node.js 20 or 22 | Optional | Installer can download a portable Node 22 for you |
Do I need to install Node.js myself?
Usually no. The one-command installer detects your environment and, if Node is missing
or too new (Node 24+ often breaks native better-sqlite3 on Windows without Visual Studio),
it downloads a portable Node 22 into %LOCALAPPDATA%\SpokesRuntime\.
Install Node yourself only if you prefer a system-wide runtime or are doing manual development.
Option A — install Node 22 LTS (recommended if you want system Node)
- Open https://nodejs.org
- Download the LTS build for Windows (prefer 22.x if available)
- Run the installer (leave “Add to PATH” enabled)
- Open a new PowerShell window and verify:
node -v
npm -v
You should see something like v22.x.x.
Avoid Node 24+ on Windows for Spokes unless you also install Visual Studio Build Tools with the Desktop development with C++ workload. Native SQLite modules need prebuilds or a C++ toolchain.
Option B — install with winget
winget install OpenJS.NodeJS.LTS --accept-package-agreements --accept-source-agreements
Then open a new terminal and check node -v.
Option C — let Spokes handle it
Skip Node entirely and run the installer below — it will fetch portable Node 22 when needed.
Optional: Git
winget install Git.Git --accept-package-agreements --accept-source-agreements
Optional: fix PowerShell script policy (only if .ps1 is blocked)
Set-ExecutionPolicy -Scope CurrentUser RemoteSigned
Or always run the installer via install.bat (it uses -ExecutionPolicy Bypass).
Quick Start
Cursor has no public API to auto-register an MCP from an
.exe, so Spokes ships an installer that writes the config for you and copies the skill.
One command (recommended)
irm https://raw.githubusercontent.com/Dente22/Spokes/main/scripts/install.ps1 | iex
Then:
- Fully restart Cursor (quit and reopen — required to reload MCP)
- Open http://127.0.0.1:3847 → click Connect GitHub
- Approve Spokes in the browser — done
No OAuth developer setup. No Personal Access Token.
Directory / Open Plugins: the repo-root
.mcp.jsonis the machine-readable MCP manifest (used by cursor.directory auto-detect). Day-to-day installs still use the one-liner above — it writes absolute paths into~/.cursor/mcp.json(required on Windows).
Double-click / release ZIP
- Download the latest release: Releases
- Extract the ZIP
- Run
install.bat - Restart Cursor → Connect GitHub
What the installer does
- Places Spokes in
%LOCALAPPDATA%\Spokes - Uses portable Node 22 when system Node is missing or Node 24+
- Runs
npm install - Merges a
"spokes"entry into%USERPROFILE%\.cursor\mcp.json(other MCPs kept) - Injects the public GitHub OAuth Client ID (
SPOKES_GITHUB_CLIENT_ID) - Copies the skill to
%USERPROFILE%\.cursor\skills\spokes\
Uninstall: uninstall.bat or scripts\uninstall.ps1 -RemoveFiles.
How it works
┌──────────────────────────────────────────────┐
You ask ───▶ │ Cursor agent + Spokes skill │
└───────────────┬──────────────────────────────┘
│ search_repos / get_file / log_step
▼
┌──────────────────────────────────────────────┐
│ Spokes MCP (local) │
│ • repo memory cards (SQLite) │
│ • GitHub API on demand │
│ • task journal │
└───────────────┬──────────────────────────────┘
│ SSE
▼
http://127.0.0.1:3847 ── Reuse flow UI
- The skill tells the agent to consult Spokes before writing non-trivial code.
search_reposfinds candidate repos;list_tree/get_filepull only what's needed.- Each reuse is recorded with
log_step. - The Reuse flow tab renders the chain live.
Web UI
Open http://127.0.0.1:3847 (starts with the MCP). The browser tab shows the Spokes favicon/logo.
| Tab | What it does |
|---|---|
| Connection | GitHub login (OAuth Device Flow) and sync settings |
| Repositories | Searchable list of repos with enable toggles |
| Reuse flow | Live task → repo → code → logic → result diagram |
MCP tools
| Tool | Purpose |
|---|---|
list_repos |
List known repo cards |
search_repos |
Find candidate repos for a task |
get_repo_card |
Full brief card for one repo |
list_tree |
Browse repo paths |
get_file |
Read a file or line range |
search_in_repo |
GitHub code search within one repo |
start_task |
Open an audit task |
log_step |
Record a reuse step |
get_task_journal |
Read task steps |
Manual install (developers)
git clone https://github.com/Dente22/Spokes.git
cd Spokes
# Prefer Node 20 or 22 on Windows
npm install
Merge mcp.json.example into ~/.cursor/mcp.json (adjust absolute paths), copy
skill/SKILL.md to ~/.cursor/skills/spokes/, restart Cursor, open the UI, Connect GitHub.
<details> <summary><b>Custom OAuth App (forks / advanced)</b></summary>
<br/>
Official Spokes builds already ship a public Device-Flow Client ID. If you fork the project, register your own GitHub OAuth App, enable Device Flow, and set:
$env:SPOKES_GITHUB_CLIENT_ID = "your_client_id"
An OAuth Client ID is public — never embed a client secret in this local application.
</details>
Sync modes
- On start (default) — refresh when Spokes/Cursor starts
- Manual — only when you click Sync
- Hourly — while the process is running
Privacy
- Runs only on your machine
- Token stored locally in SQLite under
~/.spokes - Code is fetched from GitHub on demand; cards keep only short navigation memory
Data location
All local state lives in ~/.spokes/:
spokes.db— cards, settings, task journalcache/— reserved for future file cache
Portable Node (if installed by Spokes): %LOCALAPPDATA%\SpokesRuntime\
Troubleshooting
npm install fails on better-sqlite3 / node-gyp / Visual Studio
You are probably on Node 24+ without C++ Build Tools. Fix:
- Re-run the official installer (it switches to portable Node 22), or
- Install Node 22 LTS from nodejs.org and re-run install, or
- Install Visual Studio Build Tools with Desktop development with C++
PowerShell blocks install.ps1
Use install.bat, or:
Set-ExecutionPolicy -Scope CurrentUser RemoteSigned
Cursor shows Cannot find module '...\src\index.ts'
Cursor does not reliably apply cwd. Paths in mcp.json must be absolute. Re-run the
installer to rewrite them.
UI shows an old/unknown API response
An older Spokes process may still be on port 3847. Restart Spokes in Cursor (Settings → MCP) and hard-refresh the page.
Favicon / logo not updating in the browser
Hard refresh (Ctrl+F5) on http://127.0.0.1:3847 after updating Spokes.
Development
npm run dev # MCP stdio + web UI
npm run dev:web # web UI only
npm run resync # re-pull repo cards
npm run typecheck
License
MIT © Dente22
<div align="center"> <br/> <img src="assets/logo.svg" alt="Spokes" width="40" height="40" /> <br/> <sub>Built to stop reinventing the wheel.</sub> </div>
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.