Skip to content
MCP server

Abe

Local ACT, BLOCK or ESCALATE check that runs before an AI agent takes an action

Runs locallyStatus and history
Install
Pricing
Open source
Platforms
CLI, self-hosted, API
Runs
Locally
Version
0.2.3
Listing updated
2 days ago

About Abe

Abe is an open-source MCP server for security. It runs locally on your machine and is started by your AI client.

Abe sits between an agent and the thing it is about to do: pay an invoice, buy servers, delete a table. You write a plain YAML policy, Abe checks the proposed action against it on your machine and answers ACT, BLOCK or ESCALATE. When several rules apply, BLOCK beats ESCALATE beats ACT, and bad input or a broken rule returns ESCALATE rather than letting the action through.

Policies support 11 operators with all, any and not, per-agent authorization, evidence requirements that report which fields are missing, risk and irreversibility levels, and action classes that always need judgment. Every decision produces a hash-linked record naming the matched rules and policy hash, optionally signed with Ed25519.

The MCP entry point is abe mcp, which exposes a fjp_check_action tool. The same engine is also available as a Python or TypeScript library, a Claude Agent SDK hook, a local HTTP sidecar and a Docker image. By default nothing leaves the machine; escalations can optionally go to the Flow resolver with an API key.

Install

Run in a terminal

claude mcp add abe --env ABE_POLICY='<ABE_POLICY>' --env FLOW_API_KEY='<FLOW_API_KEY>' -- uvx abe-ai

Add to claude_desktop_config.json

{
  "mcpServers": {
    "abe": {
      "command": "uvx",
      "args": [
        "abe-ai"
      ],
      "env": {
        "ABE_POLICY": "<ABE_POLICY>",
        "FLOW_API_KEY": "<FLOW_API_KEY>"
      }
    }
  }
}

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "abe": {
      "command": "uvx",
      "args": [
        "abe-ai"
      ],
      "env": {
        "ABE_POLICY": "<ABE_POLICY>",
        "FLOW_API_KEY": "<FLOW_API_KEY>"
      }
    }
  }
}

Add to .vscode/mcp.json

{
  "servers": {
    "abe": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "abe-ai"
      ],
      "env": {
        "ABE_POLICY": "<ABE_POLICY>",
        "FLOW_API_KEY": "<FLOW_API_KEY>"
      }
    }
  }
}

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

{
  "mcpServers": {
    "abe": {
      "command": "uvx",
      "args": [
        "abe-ai"
      ],
      "env": {
        "ABE_POLICY": "<ABE_POLICY>",
        "FLOW_API_KEY": "<FLOW_API_KEY>"
      }
    }
  }
}

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

{
  "mcpServers": {
    "abe": {
      "command": "uvx",
      "args": [
        "abe-ai"
      ],
      "env": {
        "ABE_POLICY": "<ABE_POLICY>",
        "FLOW_API_KEY": "<FLOW_API_KEY>"
      }
    }
  }
}

Configuration

  • ABE_POLICYAbsolute path to your abe-policy.yaml (create one with: uvx abe-ai init)Required
  • ABE_RESOLVERSet to flow to send escalations to Flow (needs FLOW_API_KEY). Default: none, nothing leaves your machine.
  • FLOW_API_KEYFlow Resolver key (fjp_live_...), only with ABE_RESOLVER=flow. https://fjp.flowinfo.co/credits

Tools this server provides (1)

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

fjp_check_action
Checks a proposed agent action against your policy and returns ACT, BLOCK or ESCALATE.

What it does

Three-way verdict
Returns ACT, BLOCK or ESCALATE with a reason code.
YAML policies
Rules, authorization, evidence needs and risk levels live in one readable file.
Fails closed
Unmatched actions, bad input and broken rules escalate instead of running.
Tamper-evident records
Each decision gets a hash-verified record, optionally Ed25519 signed.
Shadow mode
Runs beside your current approval process to show where it would agree.
Optional Flow resolver
Sends only escalations, redacted, to Flow for a judgment.

Good for

  • Stop an agent from approving a purchase above a set limit
  • Block force-pushes or DROP TABLE in production
  • Keep an audit trail of why an agent acted
  • Trial a rule set in shadow mode before enforcing it

Pricing

Open source

Apache-2.0; optional Flow resolver about $0.03-$0.05 per escalation

Prices change. Check fjp.flowinfo.co before you buy.

Abe FAQ

Does Abe need an API key or network?

No. Checks run locally with no account, key, network or model. Only the optional Flow resolver needs FLOW_API_KEY.

What happens when no rule matches?

The default policy escalates, so the agent holds and asks a person or resolver.

What licence is it?

The README states Apache-2.0.

What does Flow cost?

The README gives about $0.03 to $0.05 per escalated action, and only when you turn the resolver on.

Is Abe free?

Yes. Abe is open source, so you can use and self-host it for free.

How do I add Abe to Claude Code?

Run this command in your terminal: claude mcp add abe --env ABE_POLICY='<ABE_POLICY>' --env FLOW_API_KEY='<FLOW_API_KEY>' -- uvx abe-ai

What are the best alternatives to Abe?

The closest alternatives to Abe are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.

Alternatives to Abe

  1. REA

    MCP server

    CLI and MCP server that lets coding agents reverse engineer apps and binaries

    Runs locally
    • Open source
    • 464K GitHub stars
    • Developer tools
  2. Codebase Memory

    MCP server

    Local MCP server that builds a code knowledge graph for AI coding agents

    Runs locally
    • Open source
    • 48.7K GitHub stars
    • Knowledge and memory
  3. GitHub

    MCP server

    Official MCP server connecting AI assistants to GitHub repos, issues and workflows

    Not checked yet
    • Remote
    • Open source
    • 34.5K GitHub stars
    • Developer tools
  4. MCP toolkit giving coding agents symbol-level code retrieval and editing

    Runs locally
    • Open source
    • 31.5K GitHub stars
    • Developer tools
  5. XERJ

    MCP server

    Local search engine for AI agents with cited answers over your code and docs

    Runs locally
    • Open source
    • 19.2K GitHub stars
    • Knowledge and memory
  6. Ouroboros

    MCP server

    Spec-first layer that pins requirements and verifies results of AI coding agent work

    Runs locally
    • Open source
    • 6.4K GitHub stars
    • Developer tools