Context Engine

⚡ Context Engine

Persistent memory for AI coding agents.

A local-first MCP server for carrying useful context, decisions, and project history across AI sessions.

Quick start · Features · Tools · How it works · Configuration · Contributing

The problem

AI coding sessions are powerful—but context often disappears between conversations. Decisions get repeated, project history gets lost, and developers spend time re-explaining the same codebase.

Context Engine gives compatible AI clients a persistent, queryable memory layer. It is designed to keep useful project context available across sessions while giving developers ways to inspect, search, organize, and correct that memory.

Think of it as a memory layer for your AI workflow—not another chat interface.

✨ Why Context Engine?

🧠 Persistent context

Preserve important project knowledge so a new session can pick up where the previous one left off.

🔎 Targeted recall

Search stored context and decisions instead of relying only on the current conversation.

🌿 Branches & history

Separate experimental work, organize memory by branches, and support merge or rollback workflows.

🛠️ Human override

Inspect the memory graph and correct or prune inaccurate stored information when needed.

🧹 Context curation

Designed to prioritize durable decisions and useful context over transient debugging noise.

🔌 MCP integration

Expose memory capabilities to AI applications through the Model Context Protocol.

🚀 Quick start

Requirements

• Node.js 18 or later
• An MCP-compatible client
• Any additional credentials required by the features you enable

Use with Claude Desktop

1. Install Node.js if it is not already installed.
2. Open Claude Desktop's MCP configuration file:
   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json
3. Add the following server entry:

{
  "mcpServers": {
    "context-engine": {
      "command": "npx",
      "args": [
        "-y",
        "-p",
        "@sham-sundar-senthil-kumar/mcp-server@0.1.4",
        "context-engine-mcp"
      ]
    }
  }
}

If your configuration already has an mcpServers object, add only the "context-engine" entry inside it—don't overwrite your existing servers.

1. Save the file and fully restart Claude Desktop.
2. Open the MCP tools panel and check that Context Engine is available.

Tip: The first launch downloads the package through npm, so it may take a little longer than later launches.

Run it directly

You can also start the published server from a terminal:

npx -y -p @sham-sundar-senthil-kumar/mcp-server@0.1.4 context-engine-mcp

This starts an MCP server over standard input/output. It is intended to be launched by an MCP client; running it manually without sending MCP protocol messages will not open a chat UI.

🧰 Available tools

Context Engine exposes memory-management tools over MCP. The names below reflect the current tool interface; exact arguments and behavior are defined by the server implementation.

Example workflow

1. Start a new AI session and retrieve relevant context with gather_context.
2. Search for a past design decision with recall.
3. Record a completed architectural decision with commit.
4. Create a branch before exploring a risky refactor.
5. Review or correct saved context using get_context_tree and override_memory_node.

Tool availability and required parameters depend on the server version and configuration. Consult the tool schemas exposed by your MCP client.

⚙️ How it works

flowchart LR
    A["AI conversation"] --> B["Context Engine MCP server"]
    B --> C["Capture and curate useful context"]
    C --> D[("Local memory store")]
    D --> E["Recall and inspect"]
    E --> F["AI assistant uses relevant context"]
    F --> G["Commit, branch, merge, or correct"]
    G --> D

The goal is a hybrid-control workflow: make useful memory accessible by default, while preserving explicit tools for inspecting and changing what has been stored.

🔐 Privacy & data

Context Engine is designed around local-first use. However, local-first does not automatically mean every operation is offline or that no data ever leaves your device.

• The MCP server runs in the environment where the client launches it.
• Review the configured providers, API keys, and integrations before using sensitive project data.
• Check the implementation and configuration to understand where memory is stored and whether any enabled model or embedding provider receives data.
• Treat stored context as project data: protect your machine, backups, and credentials accordingly.

Before using this with confidential source code or personal information, review the data flow and provider settings for your specific setup.

📦 Packages & links

🧑‍💻 Development

Clone the repository:

git clone https://github.com/shamsundarsk/contextengine-.git
cd contextengine-
npm install

Build the MCP server:

npm run build -w packages/mcp-server

If you're working on the core package as well, build it using the workspace script defined by the repository. The project uses a workspace layout, so run commands from the repository root unless a package script says otherwise.

🗺️ Roadmap

Potential areas for continued development:

• [ ] Expand integration and client setup guides
• [ ] Add end-to-end tests for core memory workflows
• [ ] Document configuration and storage behavior in detail
• [ ] Add troubleshooting guides and example workflows
• [ ] Improve security, dependency auditing, and release automation

🤝 Contributing

Ideas, bug reports, documentation improvements, and pull requests are welcome.

1. Open an issue describing the use case or bug.
2. Keep changes focused and include reproduction steps for bug reports.
3. Add or update tests where appropriate.
4. Submit a pull request with a clear summary.

Please avoid including real API keys, private project data, or other secrets in issues and pull requests.

📄 License

Check the repository's LICENSE file for the applicable license.

Less re-explaining. More building.

Context Engine — persistent memory for the work behind your code.