MCP Think Tank

MCP Think Tank is a dedicated server for the Model Context Protocol that helps AI assistants reason more clearly, remember context across sessions, and use tools responsibly. It provides a structured think space, a persistent knowledge graph, and safeguards to manage tool usage and performance for complex workflows.

How to use

You connect your MCP client (such as Cursor or Claude @Web) to MCP Think Tank to give the AI a structured space for thinking, memory, and tool use. Start the server, configure your client to point at it, and then interact as you would with any MCP-enabled agent. Use the think space for structured reasoning when faced with complex decisions, query memory to reference past observations, and plan work with the task tools. The web research tools can fetch current information to inform reasoning, while safeguards manage how many tool calls happen in a single user interaction.

How to install

Prerequisites before you begin: install Node.js (version 18 or newer is recommended). You may also want a modern npm version to ensure smooth package management.

Choose one of the following install methods based on your preference. Each method keeps MCP Think Tank ready to run in your environment.

NPX (Recommended) use within your MCP client or terminal: run the latest version without a global installation.

npx mcp-think-tank@2.0.7

Global installation for a persistent CLI tool you can run from anywhere.

npm install -g mcp-think-tank
mcp-think-tank

If you prefer an explicit client configuration, you can add MCP Think Tank to your Cursor setup using the following example in your mcp.json file.

{
  "mcpServers": {
    "think-tank": {
      "command": "npx",
      "args": ["-y", "mcp-think-tank@2.0.7"],
      "type": "streamable-http",
      "env": {
        "MEMORY_PATH": "/absolute/path/to/your/project/memory.jsonl",
        "EXA_API_KEY": "your-exa-api-key-here"
      }
    }
  }
}

Available tools

think

Structured thinking tool used for formal reasoning and self-reflection with optional self-reflection flow.

memory_query

Query the knowledge graph by time, tags, or keywords to retrieve observations and entities.

upsert_entities

Create or update entities in the knowledge graph with new observations, relationships, and metadata.

add_observations

Append new facts to existing entities in the knowledge graph.

create_relations

Connect related concepts by creating relationships between entities in the graph.

search_nodes

Search the knowledge graph for specific concepts or nodes by keyword.

open_nodes

Retrieve full details of known entities from the knowledge graph.

plan_tasks

Create and organize tasks with priorities and dependencies.

list_tasks

Filter and view tasks by status or priority.

next_task

Identify and activate the highest priority task.

complete_task

Mark tasks as completed and record the outcome.

update_tasks

Update existing tasks with new information.

exa_search

Perform web searches using the Exa API for up-to-date information.

exa_answer

Return concise, sourced factual answers from web results.