TPC Server

TPC Server is a Node.js/Express API designed for AI-human collaboration, offering persistent storage for thoughts and plans and an MCP server interface for AI clients via a standard input/output transport. It enables seamless querying, creation, and organization of plans and thoughts, with an emphasis on memory context and collaborative workflows.

How to use

You can integrate an MCP client with the TPC Server using the built-in MCP server that communicates over stdio. Start the MCP server, then connect your client to the runtime to exchange commands for listing, creating, updating, and searching thoughts and plans. The server exposes resources for plans, thoughts, and context, and includes endpoint-like capabilities that your MCP client can leverage to manage and retrieve data.

How to install

Prerequisites you need before installing the server:

• Node.js (LTS version) installed on your machine
• npm package manager available in your environment

Steps to install and prepare the server:

# 1) Install dependencies
npm install

# 2) Start the Express API server
node server.js

# 3) The web UI will be available at http://localhost:3000/index.html

To run the MCP server that communicates over stdio, use the following command shown by the project setup:

npm run mcp

Additional setup notes

• The server uses a SQLite database for persistence at data/tpc.db. It supports migrations to keep the schema up to date as features evolve.

• The MCP server is designed for AI clients to connect via stdio, with a predefined set of tools and resources that your MCP client can call or navigate.

MCP server configuration and available tools

The MCP interface exposes a set of tools that your AI client can invoke to work with plans and thoughts. You can list, retrieve, create, update, and search these entities, as well as access contextual data.

Security and usage notes

Keep your environment secure and manage access to the MCP interface according to your deployment needs. Use proper authentication when exposing the API and ensure you follow local security practices for Node.js applications.

Troubleshooting tips

If you encounter startup issues, ensure Node.js and npm are installed, dependencies are installed (npm install), and there are no port conflicts for the HTTP server. For MCP connection problems, verify that your client is configured to read from standard input and write to standard output as expected by the stdio transport.

Notes on usage with the MCP client

• Use the MCP client to perform operations such as listing all plans and thoughts, creating new entries, updating plan statuses, and performing full-text searches across plans and thoughts.

• The MCP interface includes resources for plans, thoughts, and context to support memory and retrieval in agent workflows.

Tooling and endpoints overview

Available tools include operations to list, fetch, create, update, and search content. The system supports tagging, filtering by tags, and contextual search to enhance agent memory and collaboration.

Available tools

list_plans

Retrieve all plans from the database, enabling your MCP client to display or process plan entries.

get_plan

Fetch a single plan by its identifier, returning details such as title, description, status, and tags.

create_plan

Create a new plan entry with a title, description, and optional metadata like tags and context.

update_plan

Update attributes of a plan, such as its status or description.

list_thoughts

Retrieve all thoughts stored in memory for quick review or processing by an agent.

create_thought

Add a new thought with content and optional metadata to the persistent store.

search_thoughts

Perform a full-text search across thoughts to locate relevant ideas or notes.

get_context

Access aggregated contextual data that supports agent memory and decision making.