BloodHound MCP

A Model Context Protocol (MCP) server that connects LLMs to BloodHound Community Edition and BloodHound Enterprise. Ask questions in natural language, get attack path analysis, run Cypher queries, and explore Active Directory, Azure/Entra ID, and OpenGraph environments — all from your AI assistant.

Demo

Watch the demonstration video

How It Works

The server exposes BloodHound CE's REST API and Neo4j graph through a set of 13 composite MCP tools, 10 reference resources, and a system prompt tuned for offensive security analysis.

Composite Tools

Each tool uses an `info_type` parameter to select what data is returned, keeping the tool surface small and token-efficient:

Resources

Reference material the LLM loads on demand — no extra API calls:

System Prompt

The `bloodhound_assistant` prompt includes behavioral rules that guide the LLM:

• Load the offensive query library before writing Cypher for any attack scenario
• Never draw privilege conclusions without checking group memberships and `admincount`
• Respect BloodHound's property naming conventions (`hasspn`, `enabled`, `admincount` — all lowercase)
• Handle uppercase name storage (`DOMAIN ADMINS@CORP.LOCAL`) correctly in filters
• Follow proper DCSync and GPO edge traversal patterns

Prerequisites

• Python 3.11+
• uv
• BloodHound Community Edition instance with data loaded
• BloodHound API credentials (Token ID + Token Key)

Installation

Run the MCP server directly from Git without cloning it first:

export BLOODHOUND_DOMAIN=your-bloodhound-instance.domain.com
export BLOODHOUND_TOKEN_ID=your-token-id
export BLOODHOUND_TOKEN_KEY=your-token-key
uvx --from git+https://github.com/mwnickerson/bloodhound_mcp bloodhound-mcp

For a reproducible integration, append a commit after the repository URL, for

example `git+https://github.com/mwnickerson/bloodhound_mcp@`.

An `uvx` installation does not read the `.env` from a separate checkout, so

the process that launches the MCP client must provide the credential variables.

For development, clone the repository and install its environment:

git clone https://github.com/mwnickerson/bloodhound_mcp.git
cd bloodhound-mcp
uv sync

Create a `.env` file in the project root:

BLOODHOUND_DOMAIN=your-bloodhound-instance.domain.com
BLOODHOUND_TOKEN_ID=your-token-id
BLOODHOUND_TOKEN_KEY=your-token-key

At startup, the MCP server makes a signed, read-only request to

`/api/v2/self`. Invalid credentials, connectivity failures, and TLS failures

stop the server before it accepts MCP tool calls. The startup request times out

after 10 seconds.

The server defaults to `https` on port `443`. Override if needed:

BLOODHOUND_PORT=8080
BLOODHOUND_SCHEME=http

TLS certificate verification remains enabled when no additional setting is

provided. For a trusted lab deployment that uses a self-signed certificate,

verification can be explicitly disabled:

BLOODHOUND_VERIFY_TLS=false

Disabling verification weakens transport security and logs a warning. Do not

use this option on untrusted networks.

Configuration

Claude Desktop

Add to `claude_desktop_config.json`:

{
  "mcpServers": {
    "bloodhound_mcp": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/mwnickerson/bloodhound_mcp",
        "bloodhound-mcp"
      ]
    }
  }
}

Claude Code

Add to `~/.claude/mcp.json`:

{
  "mcpServers": {
    "bloodhound_mcp": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/mwnickerson/bloodhound_mcp",
        "bloodhound-mcp"
      ]
    }
  }
}

OpenAI Codex CLI

Add to `~/.codex/config.toml` (or `.codex/config.toml` for project-scoped config):

[mcp_servers.bloodhound_mcp]
command = "uvx"
args = ["--from", "git+https://github.com/mwnickerson/bloodhound_mcp", "bloodhound-mcp"]

The server inherits credentials from the process that launches Codex. To keep

them in the MCP configuration instead, pass them explicitly:

[mcp_servers.bloodhound_mcp]
command = "uvx"
args = ["--from", "git+https://github.com/mwnickerson/bloodhound_mcp", "bloodhound-mcp"]

[mcp_servers.bloodhound_mcp.env]
BLOODHOUND_DOMAIN = "your-bloodhound-instance.domain.com"
BLOODHOUND_TOKEN_ID = "your-token-id"
BLOODHOUND_TOKEN_KEY = "your-token-key"

MCP Inspector

• Command: `uvx`
• Args: `--from git+https://github.com/mwnickerson/bloodhound_mcp bloodhound-mcp`

BloodHound API Token

1. Log into BloodHound CE or BloodHound Enterprise

2. Navigate to Administration → API Tokens

3. Create a new token and copy the Token ID and Token Key into your `.env`

Usage

Example Queries

Reconnaissance:

What domains are in BloodHound?
Show me all Domain Admins in CORP.LOCAL
Find all kerberoastable users
Which computers have unconstrained delegation?

User and Group Analysis:

What admin rights does jsmith@corp.local have?
Show me all sessions for the administrator account
What groups is this user a member of?
Who controls the IT ADMINS group?

Attack Path Analysis:

Find the shortest path from jsmith@corp.local to Domain Admins
Who has DCSync rights in the domain?
Show me all GPO abuse paths
Find ADCS ESC1 paths in the domain

Custom Cypher:

Run a Cypher query to find all users with SPN set and admincount=1
Find all computers where DOMAIN USERS can RDP

Collection Uploads:

Upload this SharpHound ZIP from /tmp/sharphound.zip into BloodHound
Upload these base64-encoded SharpHound ZIP bytes as sharphound.zip
Start an upload job, upload these base64 JSON bytes as users.json, then end the job

Agents that already hold a SharpHound or AzureHound collection in memory should

base64-encode the collection bytes and call:

file_upload(
    info_type="upload_bytes",
    file_name="sharphound.zip",
    file_bytes_base64=""
)

For multi-file jobs, call `start_job`, then `upload_bytes_to_job` for each

base64 payload, then `end_job`.

OpenGraph Support

BloodHound 8.0+ supports custom node types via OpenGraph, letting you model non-AD infrastructure (cloud resources, databases, custom assets) in the same graph as Active Directory.

The `custom_nodes` tool handles legacy CRUD operations on node type display configurations through `/api/v2/custom-nodes`. For BloodHound v9.0.0+ instances with OpenGraph extension management enabled, the same composite tool also supports `/api/v2/extensions` and `/api/v2/extensions-edges` via `extension_list`, `extension_upsert`, `extension_delete`, and `extension_edges`.

Use the `bloodhound://opengraph/guide` and `bloodhound://opengraph/examples` resources for schema design and Cypher patterns. For structured OpenGraph schemas, upsert the extension schema first, then ingest collection data with `file_upload`.

> Requires BloodHound Enterprise or BloodHound CE 8.0 or later.

> OpenGraph extension management requires BloodHound 9.0.0+ and the corresponding feature flag to be enabled.

Security Considerations

BloodHound data processed through this tool is transmitted to your LLM provider's servers. Do not use this with production AD data unless you have assessed that risk.

Recommended use cases:

• Lab environments (GOAD, DetectionLab, custom ranges)
• Training and certification prep
• Research and tool development
• Non-production domain analysis

Best practices:

• Rotate BloodHound API tokens regularly
• Use a read-only API token where possible
• Consider a local LLM bridge for sensitive environments

Testing

# Full test suite
uv run pytest

# Specific modules
uv run pytest tests/test_main_mcp_tools.py -v
uv run pytest tests/test_bloodhound_api.py -v

# Integration tests (requires a live BloodHound instance)
BLOODHOUND_INTEGRATION_TESTS=1 uv run pytest tests/test_integration.py -v

Roadmap

• [ ] Direct Neo4j access mode (bypass REST API for complex graph traversal)
• [ ] Enhanced Azure/Entra ID tooling
• [ ] Improved ADCS attack path coverage
• [ ] Additional OpenGraph examples and templates

Contributing

Contributions are welcome. Open an issue to discuss significant changes before submitting a PR.

1. Fork the repo

2. Create a feature branch

3. Add tests for new functionality

4. Run `uv run pytest` and confirm everything passes

5. Submit a pull request

Acknowledgments

• SpecterOps for BloodHound Community Edition
• Orange Cyberdefense for GOAD (used for testing)
• @jlowin for FastMCP
• @xpn for MCP inspiration via the Mythic MCP project

License

GNU General Public License v3.0 — see LICENSE for details.