Skip to content
MCP server

Imperal MCP

Persistent memory, code map, undo and fact checking for AI coding agents

Runs locallyStatus and history
Install
Pricing
Freemium
Platforms
CLI, macOS, Linux
Runs
Locally
Version
0.9.20
Listing updated
today

Similar MCP servers

About Imperal MCP

Imperal MCP is a freemium MCP server for developer tools. It runs locally on your machine and is started by your AI client.

Imperal MCP wraps an existing coding agent with a server that keeps a record of what happened. Its core claim is that the agent's words are checked against what the kernel actually saw: names, files and sizes in an answer are matched against what the task read and ran, and an answer naming a function nobody opened is sent back. A plan item closes only on proof, such as a diff or a check that exited 0.

The tools listed include frame (the task, what is proven and the code around your focus), read_symbol (only the lines of one function), impact_of_change (every caller before an edit), checkpoint (undo in one call without touching .git), recall (history across sessions and tools) and token_savings, which measures your own numbers. The README says it does not make the model smarter; it stops the agent passing off what it never saw.

The code tools work offline with no account. Memory, the frame and the fact check need an Imperal Cloud account; sign-up is free and plans are on its pricing page. It installs with pipx and supports Claude Code, Cursor, Codex, Gemini CLI, Windsurf, Claude Desktop, Cline and Goose.

Install

Run in a terminal

claude mcp add imperal-mcp --env IMPERAL_TOKEN='<IMPERAL_TOKEN>' -- uvx imperal-mcp

Add to claude_desktop_config.json

{
  "mcpServers": {
    "imperal-mcp": {
      "command": "uvx",
      "args": [
        "imperal-mcp"
      ],
      "env": {
        "IMPERAL_TOKEN": "<IMPERAL_TOKEN>"
      }
    }
  }
}

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "imperal-mcp": {
      "command": "uvx",
      "args": [
        "imperal-mcp"
      ],
      "env": {
        "IMPERAL_TOKEN": "<IMPERAL_TOKEN>"
      }
    }
  }
}

Add to .vscode/mcp.json

{
  "servers": {
    "imperal-mcp": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "imperal-mcp"
      ],
      "env": {
        "IMPERAL_TOKEN": "<IMPERAL_TOKEN>"
      }
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "imperal-mcp": {
      "command": "uvx",
      "args": [
        "imperal-mcp"
      ],
      "env": {
        "IMPERAL_TOKEN": "<IMPERAL_TOKEN>"
      }
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "imperal-mcp": {
      "command": "uvx",
      "args": [
        "imperal-mcp"
      ],
      "env": {
        "IMPERAL_TOKEN": "<IMPERAL_TOKEN>"
      }
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Configuration

  • IMPERAL_API_URLOverride the Imperal API base URL (default: https://auth.imperal.io). Leave unset for production.
  • IMPERAL_TOKENOptional service token for CI or headless environments. When set, skips browser login. Obtain from panel.imperal.io → Developer → Access tokens.

Tools this server provides (8)

Taken from the project documentation; the running server may differ.

login
Sign in to Imperal Cloud with a link and code
frame
The task, what is proven and the code around your focus
read_symbol
Read only the lines of a given symbol
impact_of_change
List every caller a change reaches
checkpoint
Create an undo point
task
Define a goal and items that close on proof
recall
Recall history across sessions
token_savings
Measure tokens saved in a session

What it does

Cross-session memory
History is shared across terminals, IDEs, the web panel and Telegram.
Symbol-level reading
Reads the few lines of a function rather than a whole file.
Blast-radius check
Lists every caller and file a change reaches before the edit.
Invisible checkpoints
Undo any change in one call without touching your git history.
Fact check
Names, files and numbers in an answer are matched against what was really read and run.
Proof-based task closing
A task item closes only on a diff or a passing check the kernel saw.

Good for

  • Resume work in Claude Code that began in Cursor
  • Check who calls a shared helper before editing it
  • Roll back a failed refactor in one call
  • Catch an agent claiming it updated a function that does not exist
  • Measure how many tokens symbol-level reads save

Pricing

Freemium

Code tools free offline; cloud plans and credits on the vendor pricing page

Imperal MCP FAQ

How do I install it?

pipx install "imperal-mcp[code]", then imperal-mcp login and imperal-mcp setup.

Does it need an account?

The code tools work offline without one. Memory, the frame and the fact check use an Imperal Cloud account.

Which agents does it support?

Claude Code, Cursor, OpenAI Codex, Gemini CLI, Windsurf, Claude Desktop, Cline and Goose.

Does it make the model smarter?

No. The README says the agent can still reach a wrong conclusion from true facts; it can no longer pass off what it never saw.

Is Imperal MCP free?

Imperal MCP has a free plan, with paid plans for more usage or features.

How do I add Imperal MCP to Claude Code?

Run this command in your terminal: claude mcp add imperal-mcp --env IMPERAL_TOKEN='<IMPERAL_TOKEN>' -- uvx imperal-mcp

What are the best alternatives to Imperal MCP?

The closest alternatives to Imperal MCP are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.

Where this comes from

  • Description: written from the project's README.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers