Ghost Agent (Coreon MCP Execution Engine)

An MCP execution engine that decouples agent planning from tool execution: natural language in, structured tool-chain results out, runnable via CLI, HTTP API, or Telegram.

Stars
★ 747
Last updated
29d ago
License
MIT
Primary language
Shell

At a glance

Works with
Portable with changesOpenAI APIClaude Code (Partial support)
You'll need
DockerPython 3.11+OpenAI API keyShell / CLINetwork accessLocal filesystemMCP Server
Typical use
Web3 developers querying Solana balances, token metadata, and DeFi data from the command line via natural language
Main limitation
Hard dependency on the OpenAI API (your own OPENAI_API_KEY); no adapters for other LLM providers are documented

What does this agent do, and when should you use it?

The core product of the Ghost-Agent repository is Coreon-MCP-Execution-Engine, a unified runtime for structured ToolCall chains. It consists of four modules: a Planner that uses LLM-based intent recognition to turn natural language into a JSON execution plan, an Executor that runs tools sequentially (or in parallel) with retries, error recovery and logging, a Tool Registry that centrally declares each tool's name, module, function signature and schema, and Connectors offering CLI, API-server and Telegram-bot entry points. The engine ships as a Docker image with zero local dependencies and needs only MCP_LANG and OPENAI_API_KEY to start. Built-in tools cover DexScreener and Binance market data, CryptoPanic news, social metrics, and on-chain APIs, with an explicit focus on the Solana ecosystem as an AI execution layer for Web3. The README also outlines an MCP payment-signal protocol integration on Solana, with autonomous agent payments still in development.

Accepts natural-language input from CLI, Telegram, or HTTP API; the Planner performs LLM-based intent recognition and emits a JSON ToolCall chain; the Executor runs tools step-by-step or in parallel, handling retries, error recovery and logging; the Tool Registry declares each tool's name, module, function and schema so new tools plug in without changing execution logic; built-in tools include market data (DexScreener/Binance), news (CryptoPanic), social metrics (Twitter/Telegram), on-chain APIs (token metadata/holders), and custom utilities; a Response Formatter emits CLI charts, Telegram messages, or API JSON. On Solana it currently queries balances, token metadata, DeFi data, and contract calls; the new coreon-mcp branch adds payment-signal detection (PaymentRequired state), suspending execution and informing the user when a paid API is required.

  1. Web3 developers querying Solana balances, token metadata, and DeFi data from the command line via natural language
  2. Crypto community operators who want a Telegram bot for instant market and news lookups with no local setup
  3. Backend engineers embedding a Docker-native execution engine that separates LLM planning from tool execution
  4. Developers plugging custom tools (formatters, indicators) into crypto-analysis workflows without touching execution logic
  5. Researchers exploring agent payment economics on Solana via the MCP payment-signal protocol branch

How do you install or deploy this agent?

  1. Requirements: Python 3.11+ and Docker (install from docker.com, verify with docker --version). 2. Create a directory: mkdir mcp-execution-env && cd mcp-execution-env. 3. Generate the env file:
cat <<EOF > .env
MCP_LANG=EN
OPENAI_API_KEY=sk-xxxxxxxxxx

EOF
Replace sk-xxxxxxxxxx with your real OpenAI API key. 4. Pull the image: docker pull coreonmcp/coreon-mcp-execution-engine.

How do you use this agent?

CLI mode: docker run --rm -it --env-file .env coreonmcp/coreon-mcp-execution-engine start cli
API server mode (port 8080): docker run --rm -it --env-file .env -p 8080:8080 coreonmcp/coreon-mcp-execution-engine start server
Telegram bot mode: docker run --rm -it --env-file .env coreonmcp/coreon-mcp-execution-engine start telegram-bot

Issue tasks in natural language; the engine generates the ToolCall chain, executes it, and returns charts, messages, or JSON. The Claude-style MCP stdio mode is available in the alpha version, detailed in the Archon-Terminal repository.

What are this agent's strengths and limitations?

Pros
  • Clean four-layer architecture (Planner/Executor/Tool Registry/Connectors) with hot-pluggable tools
  • Docker-native delivery with zero local dependencies and three run modes (CLI/API/Telegram) out of the box
  • MCP payment-signal detection is implemented: execution suspends and notifies the user on a PaymentRequired state
  • Independently security-audited by Armors Labs with a public PASSED report
Limitations
  • Hard dependency on the OpenAI API (your own OPENAI_API_KEY); no adapters for other LLM providers are documented
  • Autonomous agent payments are still a roadmap item (coming soon); payment requires manual user action
  • Inconsistent naming across the README (Ghost-Agent / Archon / Coreon MCP) and links to multiple external repositories raise traceability cost
  • Solana integration is query-focused; PancakeSwap natural-language swaps, AI wallet assistants, and on-chain security monitoring remain unimplemented roadmap items

How does this agent compare with similar options?

The README positions the project as an MCP (Model Context Protocol) execution engine and notes that the alpha version officially supports the Claude-style MCP protocol (stdio mode) since 2025-09, making it comparable to invoking MCP tools directly within the Claude ecosystem.

Key facts side by side with the most closely related agents.

Agent Source review Stars Updated Language Full support on
Ghost Agent (Coreon MCP Execution Engine) This agent 0 · Blocked ★ 747 29d ago Shell OpenAI API
Zeroshot Verification Loop 63 · Some gaps ★ 1.9k today Rust Codex · Claude Code
zot Coding Agent 59 · Major gaps ★ 342 1d ago Go OpenAI API · Claude API
ZhikunCode 56 · Major gaps ★ 497 4d ago Java OpenAI API · Claude API

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Blocked
0/ 100 5-point scale 0.0 / 5
Trust 0/29
Reliability 0/14
Adaptability 0/18
Convention 0/18
Effectiveness 0/13
Verifiability 0/8
Why each dimension lost points
Trust0 / 29 · 0.0/5

Files contradict each other: the README self-describes 'Ghost-Agent / Coreon MCP Execution Engine', while package. is 'ralph-claude-code' (author frankbria, ISC license), tests reference ralph_loop.sh and lib/response_analyzer.sh which do not exist in the provided tree, and the MIT LICENSE holder 'Ghost Agent' conflicts with the ISC declaration. This stitched identity constitutes a fabricated-provenance red-line risk; the claimed 'Armors Labs PASSED' audit points to a different repository and cannot be substantiated from the reviewed files. All trust criteria are zero and the review is blocked.

Reliability0 / 14 · 0.0/5

The package. declares bats test entry points (tests/unit, tests/integration) that do not match the bash scripts provided; those scripts source non-existent files, so self-consistency and dependency claims cannot hold.

Adaptability0 / 18 · 0.0/5

The README claims CLI/API/Telegram modes and Solana integration, but no corresponding implementation files are present; capability boundaries and scenarios are pure assertions that cannot be checked.

Convention0 / 18 · 0.0/5

Naming is severely unstable (Ghost-Agent / Archon / Coreon-MCP-Execution-Engine / ralph-claude-code coexist), license metadata contradicts itself (MIT vs ISC), and version history and maintenance responsibility point to multiple unrelated third-party repositories.

Effectiveness0 / 13 · 0.0/5

Quick start relies on an external Docker image (coreonmcp/coreon-mcp-execution-engine); the repository contains no runnable artifact, so value and cost cannot be assessed from source and no credit may be given for unverified claims.

Verifiability0 / 8 · 0.0/5

Core claims (audit passed, MCP protocol support, Solana integration) rest entirely on external links with zero internal corroboration; fact and inference cannot be separated, and README marketing assertions conflict with actual file contents.

Risks and how to mitigate them
  • Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
  • Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
  • Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
  • Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
  • Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
  • Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
  • Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
  • Not found in source: verifiable attributionInstall from the official repo or registry and check the publisher and URL to avoid look-alike packages.
  • Red-line warning: the repository presents multiple unrelated project identities (Ghost-Agent, Archon, Coreon-MCP, ralph-claude-code), suggesting stitching or fabricated provenance; do not trust or deploy in production.
  • The claimed third-party security audit (Armors Labs 'PASSED') links to an external repository with no corroboration in source; treat as unverified until independently confirmed.
  • Installation relies entirely on an external Docker image that cannot be inspected from source and may carry arbitrary code.
  • The README instructs writing OPENAI_API_KEY in plaintext to .env with no key-protection guidance.
  • Test scripts source non-existent files, so the claimed test coverage is not credible.
Evidence confidence: Low Reviewed Sep 11, 2026 Reviewed revision 465cf43e5fa9
See the full review method →

FAQ

What credentials and costs are required?
You must supply an OpenAI API key (OPENAI_API_KEY in .env); costs for LLM calls and built-in market/news APIs are yours to bear, and some paid APIs are detected at runtime with a payment prompt.
What run modes are supported?
Three documented Docker modes: CLI, API server (port 8080), and Telegram bot, plus a Claude-style MCP stdio mode in alpha.
How is the language configured?
Set MCP_LANG to EN or ZH in .env; the README marks it as required.
How far does on-chain capability go?
Currently balance, token metadata, DeFi data, and contract-call queries on Solana; natural-language swaps, AI wallet assistants, and security monitoring are roadmap only.
Is security verified?
Yes — the Coreon MCP Execution Engine passed an independent audit by Armors Labs, with the full report published as a PDF.
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents