MCP-Bastion

Enterprise-Grade Security Middleware for the Model Context Protocol

Releases are published to npm, PyPI, and prebuilt Docker images on GitHub Container Registry (ghcr.io/vaquarkhan/mcp-bastion-proxy, ghcr.io/vaquarkhan/mcp-bastion-dashboard) — pull/run details in DOCKER.md — via GitHub Actions on tag push.

Documentation: structured paths for policy and LLM integration live in docs/README.md and docs/index.md. Community: open a GitHub Issue for bugs or gaps, a Discussion for integration questions (if enabled on the repo), or a PR for docs and examples—those help every adopter.

Hello world (minimal Bastion on code): see docs/QUICK_START.md — FastMCP helper secure_fastmcp(mcp) (wires MCPBastionMiddleware into tool dispatch), or two-line build_middleware_from_config() for full bastion.yaml policy, plus a CI validate snippet for pipeline-driven installs.

The Model Context Protocol (MCP) has rapidly become the universally accepted standard for connecting AI agents to enterprise databases and APIs. However, this connectivity introduces a massive new attack surface: unpredictable, non-deterministic agentic behavior.

MCP-Bastion is a lightweight, drop-in security middleware designed to wrap around any existing Python or TypeScript MCP server. Instead of relying on passive logging, human-in-the-loop approvals, or third-party APIs, MCP-Bastion provides an active, 100% local defense layer. It intercepts standard JSON-RPC traffic to stop threats before they cross the enterprise boundary.

Under 5ms proxy overhead. MCP-Bastion provides:

• Prompt Injection Defense: Meta PromptGuard runs locally to block adversarial payloads and jailbreaks.
• PII Redaction: Uses Microsoft Presidio to detect and mask PII before it reaches the LLM context.
• Infinite Loop Protection: Token buckets and cycle detection stop runaway agents from burning API budget.

Secure your MCP server without changing business logic.

Core Features

Zero-Click Prompt Injection Prevention

Integrates Meta's PromptGuard model locally to detect and block malicious payloads, jailbreaks, and adversarial tokenization before they reach your external tools.

PII Redaction

Microsoft Presidio scans outbound tool results and masks PII (redaction, substitution, generalization).

Infinite Loop and Denial of Wallet Protection

Implements stateful cycle detection and configurable FinOps token-bucket algorithms to automatically terminate runaway agents and prevent massive API bill overruns.

100% Local Execution (Data Privacy)

All security classification and data redaction happen entirely within the local memory space of your server. Sensitive data never leaves your enterprise network for third-party safety evaluations.

Low Latency

Drop-in middleware, under 5ms overhead.

Framework Integration

Hooks into MCP SDKs (TypeScript, Python) and FastMCP via standard middleware. No business logic changes.

Complete feature catalog

Pillar definitions: Security controls, bastion.yaml sections, and how they relate to dashboard health rows are documented in docs/PILLARS.md (canonical reference; avoids ambiguous “total pillar” counts). The same page lists extended features restored in 1.0.16+ (semantic firewall, sensitive classifier, external policy, edge auth, tool allowlist, session scope, tool metadata guard, multi-tenant, audit hash chain, pricing hooks, telemetry sinks, red team and doctor CLIs, etc.).

Deeper context: docs/SECURITY_OBSERVABILITY.md — OWASP MCP Top 10 alignment, attack scenarios, and SIEM/log integrations. Framework add-ons (LangChain, OpenAI, Bedrock, …) are listed under Framework Integrations below.

Threat prevention & content safety

Access control, integrity & abuse

FinOps & performance

Audit, metrics & alerting

Policy, packaging & developer experience

Real-Time Dashboard and Alerts

🎥 Demo (screen recording): Watch on Vimeo — overview of the dashboard and metrics (link opens the player on Vimeo).

Run the optional dashboard for a live view of requests, blocked count, PII redacted, cost, top tools, and recent alerts.

mcp-bastion dashboard --port 7000
# or: PYTHONPATH=src python dashboard/app.py

Dashboard: total requests, blocked count and %, PII redacted, cost; blocked-by-reason bars; top tools; cost by user; recent alerts — open http://localhost:7000/ while mcp-bastion dashboard is running.

• Alerts: Slack webhook and cost-threshold alerts. See dashboard/README.md.

Documentation: Use Cases, Attacks, Metrics, Tutorials

Adoption paths (start-to-finish):

Full index: docs/README.md (docs hub) · published site entry: docs/index.md · quick wrap: docs/QUICK_START.md · discovery: docs/DISCOVERY.md · contribute: CONTRIBUTING.md.

One-Line Docker

Prebuilt images (after the first publish-docker run, usually on a v* release tag):

docker pull ghcr.io/vaquarkhan/mcp-bastion-proxy:latest
docker run -p 8080:8080 ghcr.io/vaquarkhan/mcp-bastion-proxy:latest
# Dashboard (optional, port 7000):
# docker pull ghcr.io/vaquarkhan/mcp-bastion-dashboard:latest
# docker run -p 7000:7000 ghcr.io/vaquarkhan/mcp-bastion-dashboard:latest

Build locally (any revision):

docker build -t mcp-bastion/proxy .
docker run -p 8080:8080 mcp-bastion/proxy

MCP endpoint: http://localhost:8080/mcp. Use docker-compose up -d for proxy; add --profile with-dashboard for the dashboard. See DOCKER.md (includes GHCR pull commands and package links for forks: replace vaquarkhan with your org or user in image paths).

Policy-as-Code (bastion.yaml)

Single config file controls policy (see docs/PILLARS.md for pillar definitions). Copy bastion.yaml.example to bastion.yaml, then:

from mcp_bastion import build_middleware_from_config
middleware = build_middleware_from_config()

See docs/POLICY_AS_CODE.md.

Tip: set hot_reload.enabled: true in bastion.yaml to apply policy changes without restarting your MCP server when using build_middleware_from_config().

CLI for developers

mcp-bastion validate              # validate bastion.yaml
mcp-bastion serve --http 8080     # run MCP server with config
mcp-bastion dashboard --port 7000 # run metrics dashboard

See docs/CLI.md.

Continuous integration (this repository)

On every pull request and push to main, .github/workflows/ci.yml runs:

1. pip install -e ".[dev,policy,dashboard]" — install the Python package with tests, YAML policy loading, and FastAPI for dashboard tests.
2. mcp-bastion validate --config bastion.yaml.example — ensure the example policy file loads.
3. pytest --cov=mcp_bastion --cov-fail-under=92 — full Python test suite with ≥92% line coverage on src/mcp_bastion (see [tool.coverage.*] in pyproject.toml for measured paths and gates).
4. npm ci and npm test — TypeScript workspace tests.

To validate your repo’s bastion.yaml in CI without cloning MCP-Bastion, see examples/ci/README.md.

OpenTelemetry

Set OTEL_EXPORTER_OTLP_ENDPOINT to export tool-call spans to OTLP. Install optional deps: pip install mcp-bastion-python[otel]. See docs/OTEL.md.

Webhook alerts and external logging

Use Slack (slack_webhook / SLACK_WEBHOOK_URL), a generic HTTP webhook (webhook_url / BASTION_WEBHOOK_URL), or multiple URLs (alerts.webhooks in bastion.yaml). POSTs can drive PagerDuty, Microsoft Teams, Datadog Events, or any HTTP collector your SIEM exposes. Configure retry/backoff in bastion.yaml (retry_attempts, retry_backoff_seconds, retry_backoff_max_seconds, timeout_seconds).

Metrics & traces: scrape the dashboard /metrics (Prometheus) or poll /api/metrics (JSON) for Grafana, Datadog, or custom pollers. Set OTEL_EXPORTER_OTLP_ENDPOINT for traces to Jaeger, Honeycomb, AWS ADOT, etc. (pip install mcp-bastion-python[otel]). Route Python logs (including LoggingAlertSink) through Fluent Bit, Vector, or the CloudWatch agent into Splunk / Elastic / CloudWatch Logs.

See docs/SECURITY_OBSERVABILITY.md for the full integration table and OWASP MCP Top 10 alignment.

Why MCP-Bastion

• Active enforcement — Intercepts MCP tool traffic so policies (prompt injection checks, PII handling, content rules, RBAC, and more) run before tools execute and before sensitive results propagate.
• Local-first classification — PromptGuard and Presidio run in your environment; you are not required to send prompts to a third-party API for guardrail scoring.
• Stateful guardrails — Per-session rate limits, iteration caps, token budgets, and cost tracking to reduce runaway loops and unexpected spend.
• Composable integration — Use bastion.yaml with build_middleware_from_config() or wire MCPBastionMiddleware / wrapWithMcpBastion in Python or TypeScript. For a separate process in front of an upstream MCP server, use a wrapper or proxy you control; see docs/INTEGRATION_MODELS.md.

Structure

Example Files

Installation

Python

uv add mcp-bastion-python
# or
pip install mcp-bastion-python
# pinned latest
pip install mcp-bastion-python==1.0.16

Prerequisites (recommended)

• PII redaction: Presidio expects the spaCy English model. After install, run:
  python -m spacy download en_core_web_sm
  Without it, PII analysis can fail at runtime.
• Policy-as-Code (bastion.yaml): install YAML support:
  pip install mcp-bastion-python[policy]
  (adds pyyaml; otherwise you may get ImportError when loading policy files).
• PromptGuard fail-open: if the PromptGuard model fails to load or inference errors, MCP-Bastion allows the request and logs a warning. Treat this as a security degradation and fix the model/runtime before production.

The PyPI wheel ships the full mcp_bastion tree (including config, cli, otel, dashboard metrics, and alert sinks). If you use an older wheel that omits modules, upgrade to the current release.

TypeScript

npm install @mcp-bastion/core

npm

Framework Integrations

Drop-in security for your favorite LLM framework. Each package auto-installs mcp-bastion-python. Downloads for each row point to pypistats.org (trends) and pepy.tech (cumulative) for that PyPI name.

pip install mcp-bastion-langchain      # LangChain agents and tools
pip install mcp-bastion-openai         # OpenAI GPT API calls
pip install mcp-bastion-anthropic      # Anthropic Claude API calls
pip install mcp-bastion-bedrock        # AWS Bedrock runtime
pip install mcp-bastion-gemini         # Google Gemini
pip install mcp-bastion-crewai         # CrewAI agent crews
pip install mcp-bastion-llamaindex     # LlamaIndex RAG pipelines
pip install mcp-bastion-groq           # Groq inference
pip install mcp-bastion-mistral        # Mistral AI
pip install mcp-bastion-cohere         # Cohere
pip install mcp-bastion-azure          # Azure OpenAI Service
pip install mcp-bastion-vertexai       # Google Cloud Vertex AI
pip install mcp-bastion-huggingface    # Hugging Face Inference
pip install mcp-bastion-deepseek       # DeepSeek AI
pip install mcp-bastion-together       # Together AI
pip install mcp-bastion-fireworks      # Fireworks AI
pip install mcp-bastion-fastmcp       # FastMCP servers

Publish (PyPI / npm)

• PyPI: python -m build && twine upload dist/* (or use GitHub Actions on tag).
• npm: From repo root, cd packages/core && npm publish --access public (or use Trusted Publishers).
• Version is set in pyproject.toml (Python), packages/core/package.json (npm), and server.json (MCP registry). Bump before releasing.

Developer Guide

Integration examples for Python and TypeScript.

Quick Start (Python)

Add MCP-Bastion to an existing MCP server in three steps:

from mcp_bastion import MCPBastionMiddleware, compose_middleware

# 1. Create the security middleware
bastion = MCPBastionMiddleware(
    enable_prompt_guard=True,
    enable_pii_redaction=True,
    enable_rate_limit=True,
)

# 2. Compose with your middleware chain (Bastion runs first)
middleware = compose_middleware(bastion)

# 3. Pass the composed middleware to your MCP server
# (integration depends on your server framework)

Examples:

# Windows: $env:PYTHONPATH="src"; python examples/full_demo.py
# Linux/Mac: PYTHONPATH=src python examples/full_demo.py

LLM integration: See docs/LLM_INTEGRATION.md for copy-paste config for OpenAI, Claude, Gemini, Mistral, and Grok.

Enterprise validation:

PYTHONPATH=src python scripts/validate_checklist.py

See VALIDATION_CHECKLIST.md and SETUP_GUIDE.md.

Python Tutorial: FastMCP Server

FastMCP server with MCP-Bastion.

Step 1: Install dependencies

pip install mcp mcp-bastion-python

Step 2: Create your server file (server.py)

from mcp.server.fastmcp import FastMCP
from mcp_bastion import MCPBastionMiddleware, compose_middleware

# Create the MCP server
mcp = FastMCP("My Secure Server")

# Create MCP-Bastion middleware
# It intercepts tool calls and resource reads before they execute
bastion = MCPBastionMiddleware(
    enable_prompt_guard=True,   # Block malicious prompts via PromptGuard
    enable_pii_redaction=True,  # Mask PII in outgoing content
    enable_rate_limit=True,     # Cap at 15 iterations, 60s timeout
)

# Compose middleware chain (pass to your server's middleware config if supported)
middleware = compose_middleware(bastion)

# Register a tool (protected when middleware is wired into your server)
@mcp.tool()
def get_weather(city: str) -> str:
    """Get weather for a city."""
    return f"Weather in {city}: 22C, sunny"

# Resource (PII redacted)
@mcp.resource("user://profile/{user_id}")
def get_profile(user_id: str) -> str:
    """Get user profile. PII redacted."""
    return f"User {user_id}: John Doe, SSN 123-45-6789, john@example.com"

if __name__ == "__main__":
    mcp.run(transport="streamable-http")

Step 3: Run the server

python server.py

MCP-Bastion:

• Scans tool args for prompt injection
• Redacts PII from resource responses
• Blocks sessions over 15 calls or 60s

Alternative: Policy-as-Code

Use bastion.yaml instead of code. Copy bastion.yaml.example to bastion.yaml, then:

from mcp_bastion import build_middleware_from_config
middleware = build_middleware_from_config()

See docs/POLICY_AS_CODE.md and examples/server_with_config.py.

Python: Custom Rate Limits

Custom config example:

from mcp_bastion import MCPBastionMiddleware
from mcp_bastion.pillars.rate_limit import TokenBucketRateLimiter
from mcp_bastion.pillars.prompt_guard import PromptGuardEngine

# Stricter limits
rate_limiter = TokenBucketRateLimiter(
    max_iterations=10,
    timeout_seconds=30,
    token_budget=25_000,
)

# Higher threshold = fewer blocks, more risk
prompt_guard = PromptGuardEngine(threshold=0.92)

bastion = MCPBastionMiddleware(
    prompt_guard=prompt_guard,
    rate_limiter=rate_limiter,
    enable_prompt_guard=True,
    enable_pii_redaction=True,
    enable_rate_limit=True,
)

# Disable PII redaction if your data has no PII
bastion_no_pii = MCPBastionMiddleware(enable_pii_redaction=False)

Python: Custom Middleware

Extend Middleware to add logging, metrics, or custom logic:

from mcp_bastion.base import Middleware, MiddlewareContext, compose_middleware

class LoggingMiddleware(Middleware):
    async def on_message(self, context, call_next):
        result = await call_next(context)
        # log method, elapsed, etc.
        return result

middleware = compose_middleware(bastion, LoggingMiddleware())

See examples/full_demo.py for a complete example.

TypeScript: Wrap an MCP Server

Step 1: Install dependencies

npm install @modelcontextprotocol/sdk @mcp-bastion/core

Step 2: Create your server (server.ts)

import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import {
  wrapWithMcpBastion,
  wrapCallToolHandler,
} from "@mcp-bastion/core";

const server = new Server({ name: "my-mcp-server", version: "1.0.0" });

// Wrap the server with MCP-Bastion (rate limiting only by default)
// For prompt injection and PII, run the Python sidecar and set sidecarUrl
wrapWithMcpBastion(server, {
  enableRateLimit: true,
  maxIterations: 15,
  timeoutMs: 60_000,
  // Optional: enable ML features via Python sidecar
  sidecarUrl: process.env.MCP_BASTION_SIDECAR || "",
  enablePromptGuard: !!process.env.MCP_BASTION_SIDECAR,
  enablePiiRedaction: !!process.env.MCP_BASTION_SIDECAR,
});

// Register tools (handlers are automatically wrapped)
server.setRequestHandler("tools/call" as any, async (request) => {
  if (request.params?.name === "get_weather") {
    return {
      content: [{ type: "text", text: "Sunny, 22C" }],
      isError: false,
    };
  }
  throw new Error("Unknown tool");
});

async function main() {
  const transport = new StdioServerTransport();
  await server.connect(transport);
}

main();

Step 3: Run with rate limiting only

npx tsx server.ts

Step 4: Run with full ML features (Python sidecar)

For prompt injection and PII redaction, run a Python HTTP service that exposes /prompt-guard and /pii-redact endpoints (see the Python package for sidecar implementation). Then:

# Start the Python sidecar, then the TypeScript server
MCP_BASTION_SIDECAR=http://localhost:8000 npx tsx server.ts

TypeScript: Wrap Individual Handlers

Wrap specific handlers only:

import {
  wrapCallToolHandler,
  wrapReadResourceHandler,
} from "@mcp-bastion/core";
import {
  CallToolRequestSchema,
  ReadResourceRequestSchema,
} from "@modelcontextprotocol/sdk/types.js";

// Wrap only the tool handler
const safeToolHandler = wrapCallToolHandler(
  async (request) => {
    // Your tool logic
    return { content: [{ type: "text", text: "OK" }], isError: false };
  },
  { enableRateLimit: true, maxIterations: 10 }
);

// Wrap only the resource handler (for PII redaction)
const safeResourceHandler = wrapReadResourceHandler(
  async (request) => {
    const contents = await fetchResource(request.params.uri);
    return { contents };
  },
  { sidecarUrl: "http://localhost:8000", enablePiiRedaction: true }
);

server.setRequestHandler(CallToolRequestSchema, safeToolHandler);
server.setRequestHandler(ReadResourceRequestSchema, safeResourceHandler);

Configuration Reference

Error Handling

When MCP-Bastion blocks a request, it returns standard MCP/JSON-RPC errors:

# Python: exceptions
from mcp_bastion.errors import (
    PromptInjectionError,
    RateLimitExceededError,
    TokenBudgetExceededError,
    CircuitBreakerOpenError,
    ContentFilterError,
    RBACError,
    SchemaValidationError,
    ReplayAttackError,
    CostBudgetExceededError,
    SemanticFirewallError,
    ExternalPolicyDeniedError,
    SensitiveContentError,
    AuthenticationError,
    ToolNotAllowedError,
    SessionScopeExceededError,
    ToolMetadataPoisoningError,
)
import logging
logger = logging.getLogger(__name__)

try:
    result = await middleware(context, call_next)
except (
    PromptInjectionError,
    RateLimitExceededError,
    TokenBudgetExceededError,
    CircuitBreakerOpenError,
    ContentFilterError,
    RBACError,
    SchemaValidationError,
    ReplayAttackError,
    CostBudgetExceededError,
    SemanticFirewallError,
    ExternalPolicyDeniedError,
    SensitiveContentError,
    AuthenticationError,
    ToolNotAllowedError,
    SessionScopeExceededError,
    ToolMetadataPoisoningError,
) as e:
    logger.warning("blocked: %s", e.to_mcp_error())

// TypeScript: handlers return isError: true
import { logger, setLogLevel } from "@mcp-bastion/core";
setLogLevel("debug");  // optional: "debug" | "info" | "warn" | "error"
const result = await guardedHandler(request);
if (result.isError) {
  logger.error("blocked", result.content);
}

Testing

MCP Inspector:

# Start your guarded server
python server.py   # or: npx tsx server.ts

# In another terminal, launch the Inspector
npx -y @modelcontextprotocol/inspector

Connect via HTTP (http://localhost:8000/mcp) or stdio, then:

1. List tools and call one with benign arguments (should succeed)
2. Call a tool with "Ignore previous instructions" (should be blocked)
3. Trigger 16+ tool calls in one session (should hit rate limit)

Testing

# Python (PYTHONPATH=src on Windows: $env:PYTHONPATH="src")
pytest tests/ -v

# TypeScript
npm run test --workspace=@mcp-bastion/core

# Full validation checklist (build, pillars, latency)
PYTHONPATH=src python scripts/validate_checklist.py

# MCP Inspector (manual)
npx -y @modelcontextprotocol/inspector

Third-Party Components

See NOTICE for licenses. MCP-Bastion uses Meta Llama Prompt Guard 2 (Llama 4 Community License) and Microsoft Presidio. For OWASP-relevant mitigations, dependency audit, and reporting vulnerabilities, see docs/SECURITY.md.

License

MCP-Bastion is distributed under the MCP-Bastion Community and Commercial License (LICENSE).

• Free for non‑commercial use when you cite MCP-Bastion and the copyright notice (see CITATION.cff; you can list your name, team, or org as authors or as who used the software, while still including the project and repository in the credit).
• Copyright is retained. Do not remove license or copyright text, and do not republish a duplicate of the work as if it were unrelated software without meeting the License terms.
• Commercial use (as defined in the License) may still require a separate written agreement — see COMMERCIAL_LICENSE.md.

See also:

• LICENSE
• COMMERCIAL_LICENSE.md
• CITATION.cff