Eidolon MCP server
Commit Nov 8, 2026Performs privacy-focused OSINT scans to identify digital footprints, data breaches, and exposed credentials using local LLMs for report generation.
Also known as sudohnim/eidolon
- Stars
- 1
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- 0.1.0
- Last commit
- Nov 8, 2026
- License
- AGPL-3.0
What is Eidolon MCP?
Eidolon performs privacy-focused OSINT scans to identify digital footprints, data breaches, and exposed credentials using local LLMs for report generation.
Summary generated from the project's README
About Eidolon MCP server
Identifying digital footprints and data breaches through privacy-focused OSINT scans is what Eidolon enables for security-conscious individuals and researchers. By aggregating approximately 25 sources including data brokers, leaked credentials, and public records, the server generates comprehensive risk reports. It maps findings to the MITRE ATT&CK framework and flags potential AI-training exposure to help users understand their actual attack surface.
Users interact with the server through MCP clients like Claude Desktop or Claude Code using tools such as scan_target, scan_status, list_scans, and get_report. The system runs locally over stdio to ensure data stays on your machine. While it can function with minimal configuration, providing API keys for services like HIBP or DeHashed expands the scan depth. For narrative report generation, it utilizes a local LLM via Ollama, such as llama3.1:8b, ensuring that sensitive information never reaches external AI services.
Security is maintained through deterministic risk scoring and a default redaction policy for the leaked-credential dossier. Plaintext passwords are only accessible if you explicitly call the reveal_credentials tool.
What you can do with it
- 1Analyze personal digital exposureAsk the AI to scan your email address to find leaked credentials and public records.
- 2Map vulnerabilities to attack frameworksRequest a report that connects discovered data leaks to specific adversary tactics.
- 3Audit AI data privacyIdentify which online platforms might be using your personal information for model training.
- 4Retrieve redacted security credentialsExplicitly request the plaintext passwords found in breach dumps after reviewing a scan.
Written from the project's README. Every use case is checked against a quote from it.
Configuration
DEHASHED_API_KEYHIBP_API_KEYOLLAMA_HOST
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablesudohnim/eidolon is public and reachablegithub.com/sudohnim/eidolon ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)AGPL-3.0spdx.org/licenses/AGPL-3.0.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-08github.com/sudohnim/eidolon ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubEidolon
Eidolon is a privacy-first approach to finding and understanding your digital footprint. The stack uses OSINT tools to gather information based on your search parameters, then a local LLM compiles it into a report. The LLM runs on your machine, so no data ever reaches an external service or leaves your box — you own your data.
What it does
- Aggregates ~25 OSINT sources — breaches, leaked credentials, data brokers, account enumeration, public records, phone/email intel, exposed hosts — into a single risk report.
- Maps findings to MITRE ATT&CK so you see what an attacker could actually do with what's exposed.
- Flags AI-training exposure (which platforms may train on your data, and how to opt out).
- Risk scoring and the leaked-credential dossier are deterministic — built from scan state, not the LLM — so the report survives an LLM hiccup. The model only writes narrative.
- Outputs Markdown, PDF, and JSON.
MCP-native
Eidolon runs as an MCP server, so you can drive it from any MCP client (Claude Desktop, Claude Code) — scan a target, list past scans, and read reports conversationally. It runs locally over stdio; your data never leaves the box.
Tools: scan_target, scan_status, list_scans, get_report, reveal_credentials. A scan takes minutes, so scan_target returns a scan_id immediately and runs in the background — poll scan_status(scan_id) until it reports done, then get_report(scan_id). The leaked-credential dossier (plaintext passwords from breach dumps) is redacted by default and only returned when you explicitly call reveal_credentials.
Requirements
Everything below is optional — Eidolon runs with whatever you give it and tells you, per source, what it couldn't check (no token) versus what it checked and found nothing.
- Python 3.11+ and uv (required)
- Ollama for the local LLM narrative:
ollama pull llama3.1:8b— if it's down, the report is still produced deterministically (just no written narrative). - A running SpiderFoot instance — skipped if unreachable.
- API keys — see
.env.example. Each unlocks one data source and skips cleanly if absent (the report says "not checked — set X"). A scan with no keys is sparse but still runs.
New here? Start with the Setup & Configuration guide — which keys to get, what each one costs, and what it unlocks.
Quickstart
git clone https://github.com/sudohnim/eidolon && cd eidolon
uv sync
cp .env.example .env # fill in your keys
# scan yourself from the CLI
uv run eidolon --email you@example.com
# or run the MCP server (stdio)
uv run eidolon-mcp
Use from Claude Desktop
Option 1 — from a local clone (loads .env from cwd):
{
"mcpServers": {
"eidolon": {
"command": "uv",
"args": ["run", "eidolon-mcp"],
"cwd": "/absolute/path/to/eidolon"
}
}
}
Option 2 — from PyPI via uvx (recommended, no clone needed). Pass your keys in the env block:
{
"mcpServers": {
"eidolon": {
"command": "uvx",
"args": ["--from", "eidolon-osint", "eidolon-mcp"],
"env": {
"HIBP_API_KEY": "...",
"DEHASHED_API_KEY": "...",
"OLLAMA_HOST": "http://localhost:11434"
}
}
}
}
Option 3 — straight from GitHub (pre-release / latest commit):
{
"mcpServers": {
"eidolon": {
"command": "uvx",
"args": ["--from", "git+https://github.com/sudohnim/eidolon", "eidolon-mcp"],
"env": {
"HIBP_API_KEY": "...",
"DEHASHED_API_KEY": "...",
"OLLAMA_HOST": "http://localhost:11434"
}
}
}
}
Add only the keys you have — the rest skip cleanly. Then ask Claude to "scan my email"; it calls scan_target, polls scan_status, and reads the report.
How it works
A LangGraph pipeline: intake → wave 1 scans → wave 2 scans → MITRE mapping → correlation → analysis → report. Both the CLI and the MCP server call the same run_scan() core; reads go through a small repository layer. See docs/ROADMAP.md for the architecture and where it's headed (stateful history, continuous monitoring).
Intended use
Eidolon is for scanning yourself, or targets you are explicitly authorized to assess (authorized security testing, your own footprint). It surfaces real secrets, including plaintext passwords from breach dumps. Do not use it to profile or surveil people without their consent. You are responsible for complying with the terms of the data sources you configure and with applicable law.
License
AGPL-3.0. If you run a modified version as a network service, you must offer users its source.
Eidolon MCP questions
Is my data safe when using Eidolon?
Yes, it uses a privacy-first approach where a local LLM compiles reports on your machine, ensuring no data leaves your box.
What happens if I don't have all the API keys configured?
The server will still run; it simply skips the sources for which it lacks credentials and notes them as not checked in the report.
How can I see plaintext passwords from breach dumps?
Leaked credentials are redacted by default and can only be accessed by explicitly calling the reveal_credentials tool.
Does the report depend on the LLM working correctly?
No, the risk scoring and credential dossier are deterministic and built from scan state, so the report remains accurate even if the LLM fails.