Skip to main content
Glama
Correctover MCP Server

Correctover MCP Server

Official
by Correctover

Correctover — MCP Runtime Security

The runtime security layer for the MCP ecosystem.

We build the runtime security layer for AI agent ecosystems. Correctover enforces security conformance on every MCP tool call — validating LLM outputs, blocking injection attacks, preventing credential hijacking, and auto-recovering from failures at 22μs P50 latency.

License npm npm PyPI DOI


What We Do

Layer

What

Where

Runtime Verification

6-dimension output validation (structure, schema, latency, cost, identity, integrity) + auto-failover

correctover-mcp-server

Agent Governance SDK

Synchronous interceptor-based governance — fail-closed by design

correctover-ccs (PyPI)

Security Audits

MCP protocol vulnerability research — CVE-class findings across 50+ implementations

mcp-security-audits

Conformance Standard

CCS v1.0 — formal standard for agentic runtime verification

standards

Fault Taxonomy

215 fault types, 19 CVEs, 561 fault variants cataloged

Internal knowledge base


Related MCP server: Kluster.ai Verify MCP

Real Data — Not Benchmarks

We don't simulate. We collect real API responses from production MCP servers and verify them.

20,000 Verified API Traces

Dataset

Records

Size

Format

CCS 20K Verification Subset

20,000

18 MB

JSONL (download)

Collection methodology:

  • 43.6 minutes of continuous collection at 13.7 API calls/second

  • 68.88% conformance rate (13,776 conformant / 6,224 non-conformant)

  • 30% fault injection rate for stress testing

  • Every record includes: request, response, latency, validation result, fault classification

Third-party independent verification: 120,426 conformance re-calculations by @babyblueviper1 — full consistency confirmed.

The complete 20K dataset is in ./data/ — download, verify, fork, do whatever you want. No gatekeeping.


Research & Publications

CCS v1.0 — Runtime Security Conformance Standard

The first formal conformance standard defining how agent runtimes should validate tool execution results at runtime.

Key findings from 20K real traces:

  • P50 validation latency: 22μs

  • Self-heal rate: 97.4% (engine auto-retries/fails over on failed validation, then re-validates)

  • Rule coverage: 88 detection rules (64 high-confidence)

  • 561 distinct fault variants cataloged across all major LLM providers

Fault Taxonomy

We maintain a living fault taxonomy derived from real-world MCP server failures:

  • 215 distinct fault types classified across 7 severity levels

  • 19 CVE-class vulnerabilities identified across MCP implementations

  • Categories: RCE, SSRF, cloud credential hijacking, path traversal, output injection, privilege escalation


Upstream PR Contributions

We don't just report — we fix. Our contributions go directly into major agent frameworks:

PR

Framework

Status

What

ferro-labs#197

Ferro Labs

OPEN

Runtime validation integration

CrewAI#6432

CrewAI

10 commits

GuardrailProvider — runtime governance protocol

CrewAI#6411

CrewAI

Discussion

Defining runtime verification authority

agent-governance-toolkit#3347

Microsoft

Under review

Runtime threat scanner — recursive nested-arg scanning, SSRF gaps, credential redaction, path boundary fixes


Community Validation

Real researchers using our work in production:

Researcher

Framework

Contribution

@pshkv (AutoGen maintainer)

AutoGen

Adopted Required(τ)⊆Supported(τ) framework for tool governance

@humbl-dev

CrewAI

Testing two-layer governance structure

@safal207

CrewAI

Implemented GuardrailProvider based on our design (10 commits)

@babyblueviper1

Independent

120,426 independent conformance re-calculations

@Tuttotorna

PHI-OMEGA

ICLR paper collaboration on runtime verification

@XYG-LUNA

CrewAI

Idempotency analysis and interaction


MCP Server — Product

The runtime verification engine packaged as an MCP server for your AI tools.

Quick Start

npm install -g correctover-mcp-server

Add to your mcp.json (Cursor, Claude Desktop, Windsurf):

{
  "mcpServers": {
    "correctover": {
      "command": "correctover-mcp-server",
      "env": {
        "OPENAI_API_KEY": "sk-...",
        "ANTHROPIC_API_KEY": "sk-ant-..."
      }
    }
  }
}

BYOK — your keys stay on your machine. No proxy, no data collection.

How It Works

Your AI Tool (Cursor / Claude Desktop / Windsurf)
        │
        ▼
┌─────────────────────────────┐
│   Correctover MCP Server    │
│   ┌───────────────────────┐ │
│   │  6-Dim Validator      │ │
│   │  ├─ Structure         │ │
│   │  ├─ Schema            │ │
│   │  ├─ Latency           │ │
│   │  ├─ Cost              │ │
│   │  ├─ Identity          │ │
│   │  └─ Integrity         │ │
│   └───────────────────────┘ │
│          │                  │
│   ┌──────▼──────┐           │
│   │ Failover    │           │
│   │ Engine      │           │
│   └──────┬──────┘           │
└──────────┼──────────────────┘
           │
    ┌──────┼──────┬──────────┐
    ▼      ▼      ▼          ▼
 OpenAI  Anthropic  DeepSeek  Qwen ...

Supported Providers

Provider

Models

Env Variable

OpenAI

GPT-4o, GPT-4o-mini, o1

OPENAI_API_KEY

Anthropic

Claude 3.5 Sonnet, Haiku, Opus

ANTHROPIC_API_KEY

DeepSeek

DeepSeek-V3, DeepSeek-R1

DEEPSEEK_API_KEY

Moonshot/Kimi

Moonshot-v1

MOONSHOT_API_KEY

Alibaba Qwen

Qwen-Max, Qwen-Plus

DASHSCOPE_API_KEY

Groq

Llama, Mixtral

GROQ_API_KEY

Together

Llama, Mistral

TOGETHER_API_KEY

Tools

Tool

Description

chat

Verified chat — 6-dim validation + auto-failover

health

Check provider status

providers

Detailed provider configuration

stats

Session metrics: calls, pass rate, failover count

validation_history

Query recent validation results (ring buffer, 500 records)


Ecosystem Adoption

Real download numbers from public package registries (last 30 days):

Package

Registry

Monthly Downloads

correctover-mcp-server

npm

1,564

correctover

npm

1,034

correctover-ccs

npm

400

correctover

PyPI

1,436

Total

4,434/month

All organic growth — no paid promotion. CCS standard package seeing highest growth rate (tens of thousands percent increase from baseline).


CCS SDK — Agent Governance

Python SDK for embedding governance into agent frameworks. Fail-closed by design.

pip install correctover-ccs
from ccs import govern

@govern(policy="default")
def my_tool(args: dict) -> str:
    return "result"

# Governance evaluates BEFORE function runs
# If denied → PermissionError, function never executes

Supported frameworks: CrewAI, AutoGen, LangGraph/LangChain

Observer hooks (default):  governance_crash → tool EXECUTES ❌
CCS decorators (ours):     governance_crash → tool BLOCKED ✅

Security Audit Reports

We publish detailed security audits of MCP server implementations:

→ mcp-security-audits

Methodology: Source code analysis → fault injection → runtime verification → CVE classification

Findings to date: 506 security findings across 3 major repositories, 5 vulnerability types confirmed cross-repo.


Resource

Link

CCS Standard (paper)

DOI: 10.5281/zenodo.21234580

CCS Standard (GitHub)

Correctover/standards

MCP Server (npm)

correctover-mcp-server

CCS SDK (PyPI)

correctover

Security Audits

Correctover/mcp-security-audits

Agent Governance

Correctover/agent-governance-toolkit (fork with PRs)

Glama

correctover/mcp-server

Protocol Spec

standards/docs/RFC-001

Website

correctover.com


Contact

Security reports: wangguigui@correctover.com
BD / Enterprise: wangguigui@correctover.com
GitHub: @Correctover

License

Apache 2.0 © Correctover

Available Tools

5 tools
chatA
Read-only

Send a chat message to an LLM with automatic output verification. Routes through the best available provider, validates the response across 6 dimensions (structure, schema, latency, cost, identity, integrity), and auto-heals on failure by retrying or failing over to another provider. Returns the response text plus a validation report showing which dimensions passed or failed.

Inputs: messages (required), model, provider, max_tokens, +2 more.

healthA
Read-onlyIdempotent

Check health and availability of all configured LLM providers. Returns a list of active providers with their default models and session statistics. Call this first to verify your configuration before using the chat tool.

providersA
Read-onlyIdempotent

List all supported LLM providers with their configuration details, default models, base URLs, and current status. Use this to see which providers are available, what model each uses by default, and whether custom base URLs are configured for proxy or mirror setups.

statsA
Read-onlyIdempotent

Show Correctover session statistics including total API calls, validation pass rate, failover count, active providers, and server version. Use this after a working session to review reliability metrics and see how many self-healing events occurred.

validation_historyA
Read-onlyIdempotent

Query recent validation results with pagination. Returns the most recent LLM call validation records, newest first. Each record includes provider, model, latency, pass/fail status, validation score, and failure reasons. Use limit and offset to paginate through results. Default returns the 20 most recent records. Maximum 100 per page. The buffer holds up to 500 records; older entries are automatically overwritten.

Inputs: limit, offset.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.3/5.0

    Scored across 5 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose: chat for sending validated messages, health for checking provider availability, providers for listing all supported providers, stats for session statistics, and validation_history for reviewing past validation results. There is no overlap in functionality.

    Naming Consistency5/5

    All tool names follow a consistent pattern: single words or compound words with underscores (e.g., validation_history). They are all lowercase and descriptive of their function, making them easy to understand and predict.

    Tool Count5/5

    With 5 tools, the server is well-scoped for its purpose of managing LLM chat with validation and monitoring. Each tool covers a core aspect without being overly specialized or too sparse.

    Completeness5/5

    The tool surface covers the full lifecycle of interacting with LLM providers: checking configuration (health, providers), sending messages (chat), reviewing performance (stats), and auditing (validation_history). No obvious gaps are present for the stated purpose.

    Maintenance

    ActivityNo data
    ResponsivenessSyncing

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      B
      quality
      F
      maintenance
      A lightweight bridge that wraps OpenAI's built-in tools (like web search and code interpreter) as Model Context Protocol servers, enabling their use with Claude and other MCP-compatible models.
      4
      12
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Provides real-time content security for large language models by identifying and intercepting risks across compliance, ethics, and safety dimensions. It enables secure input and output monitoring through a customizable policy engine using an SSE-based interface.
      1
      MIT