Contextforge MCP

What alfredoizdev/contextforge-mcp MCP server does

The alfredoizdev/contextforge-mcp MCP server connects MCP-compatible coding assistants to long-term project memory. Agents can save decisions, architecture notes, debugging details, and other knowledge, then retrieve related information with semantic search rather than relying only on exact keywords. Stored items can be listed, deleted individually, or processed in batches.

Memory is organized through spaces and projects. The available project operations can create or remove projects, associate a local directory with a project, show the active project, and move spaces between projects. This structure is useful when one developer or team works across multiple repositories or separate areas of knowledge.

The server also includes GitHub repository operations. An agent can connect and disconnect repositories, list connected repositories, enable or disable webhook activity, import existing history, and inspect synchronized commits and pull requests. Issue tools support listing and creating issues, starting work, resolving issues by identifier or title, assigning collaborators, and asking what to work on next.

How it works

The server runs locally as an MCP process while using the ContextForge service for the account and stored data. Clients start it with Node.js through npx, and authentication is supplied with the CONTEXTFORGE_API_KEY environment variable. A free tier is available, but an API key must be generated from the ContextForge website.

From version 0.5.1 onward, the server sends startup instructions when a client supports them. Claude Desktop and Claude Code can therefore load project memory at session start and save relevant decisions without additional editor rules. Clients that do not honor those instructions can use the init command to add explicit memory, session-presence, and startup-context rules to CLAUDE.md or .cursorrules.

The README describes a lean default tool exposure to reduce context usage, with a full mode available through CONTEXTFORGE_TOOLS=full. The supplied tool set covers memory, spaces, projects, GitHub synchronization, and issue tracking.

Setup and configuration

Install-on-demand configuration uses npx -y contextforge-mcp as the MCP command. Configure the command and CONTEXTFORGE_API_KEY in the client’s MCP settings, then restart the client. The documented clients include Claude Desktop, Claude Code, Cursor, and GitHub Copilot; the project description also names Windsurf as supported.

Generate the API key at contextforge.dev under Settings and API Keys. The key starts with cf_ according to the setup instructions. Node.js 20 or newer is required. A global npm installation is optional, while npx fetches the package when the client launches it.

Running npx contextforge-mcp init is recommended when explicit project instructions are useful. It detects Claude or Cursor project markers, or can target an editor with --editor=claude, --editor=cursor, or --editor=all. Re-running the command preserves existing sections and adds missing ones.

Limitations and notes

The server depends on the external ContextForge service and an account API key; it is not a standalone local memory database. Automatic startup behavior varies by client, so editor rule generation may still be needed for clients that do not process server instructions. Git synchronization requires connecting a GitHub repository before history, commit, pull-request, or webhook operations can provide repository data.

The project is open source under the MIT license. Its package requires Node.js 20 or newer, and its free tier is available without a credit card according to the README.