MCP 服务器

Blekline MCP Server

Provides AI ingress governance by masking prompts, classifying risk, and enforcing tool policies before agent calls reach model providers or sandboxes.

README

<h3 align="center"> Run secure AI agents and code. <br /> <sub>AI ingress control plane — mask prompts, enforce MCP tool policy, and audit every agent call before models and sandboxes.</sub> </h3>

<p align="center"> <a href="docs/introduction/quick-start.md">Documentation</a> · <a href="https://github.com/Blekline/blekline-oss/issues">Report Bug</a> · <a href="https://github.com/Blekline/blekline-oss/issues">Request Feature</a> · <a href="SECURITY.md">Security</a> · <a href="https://app.blekline.com">app.blekline.com</a> </p>

Blekline is the ingress governance layer for enterprise AI adoption. Agents in Cursor, Claude Desktop, and Codex call Blekline to redact sensitive data, classify risk, and enforce workspace MCP tool policy before traffic reaches LLM providers or Daytona sandboxes.

The open-source packages in this repository cover local enforcement, MCP tools, SDK clients, and an optional Docker ingress sidecar. The managed control plane at app.blekline.com adds Azure-backed PII masking, fleet policy, investigations, and billing.

Features

Area	Capabilities
Ingress	OpenAI/Anthropic-compatible proxy paths, regional edge sidecar (Helm)
MCP	`blekline_mask_prompt`, `blekline_classify_risk`, `blekline_evaluate_tool_call`, `blekline_emit_event`
MCP proxy	Intercept downstream tool calls → allow / mask / block → forward to Daytona or other MCP
SDK	TypeScript (`@blekline/client`) and Python (`blekline-client`)
Local	Secret scan + tool policy in `@blekline/contracts` (no cloud required for dev)
Audit	Metadata-oriented events; client surface labels (cursor, claude-desktop, codex)

Platform comparison

Capability	OSS (this repo)	Cloud (app.blekline.com)
MCP server / proxy	Yes	Yes
Local tool enforce	Yes	Yes
Azure authoritative PII	—	Yes
Workspace fleet policy (SSE)	—	Yes
Investigations / billing	—	Yes
Browser extension wedge	—	Yes (proprietary)

Architecture

[ Cursor | Claude Desktop | Codex ]
           │ MCP (stdio / SSE)
           ▼
[ @blekline/mcp-server | @blekline/mcp-proxy ]
           │ HTTPS (optional)
           ▼
[ Blekline control plane — mask · enforce · audit ]
           ├──────────────► [ Model providers ]
           └──────────────► [ Daytona sandbox — approved tools only ]

Details: docs/introduction/architecture.md · Daytona stack

Packages

Runnable packages live under packages/:

Package	npm / PyPI	License
`@blekline/contracts`	(workspace)	Apache-2.0
`@blekline/client`	`@blekline/client`	Apache-2.0
`blekline-client`	PyPI (optional)	Apache-2.0
`@blekline/mcp-server`	`@blekline/mcp-server`	AGPL-3.0
`@blekline/mcp-proxy`	`@blekline/mcp-proxy`	AGPL-3.0
`ingress-proxy`	Docker image	AGPL-3.0

OpenAPI spec: packages/contracts/openapi.yaml

Client libraries

TypeScript

npm install @blekline/client @blekline/mcp-server

import { BleklineClient } from "@blekline/client";

const blekline = new BleklineClient({
  workspaceToken: process.env.BLEKLINE_WORKSPACE_TOKEN!,
  metadata: { clientSurface: "sdk" },
});

const { maskedText, decision } = await blekline.mask({
  text: "Contact alice@corp.com — key AKIAIOSFODNN7EXAMPLE",
  platform: "MyAgent",
});

const tool = await blekline.enforceToolCall({
  toolName: "run_shell",
  arguments: { cmd: "curl https://api.internal/deploy" },
});

Python

pip install blekline-client

import os
from blekline_client import BleklineClient

client = BleklineClient(workspace_token=os.environ["BLEKLINE_WORKSPACE_TOKEN"])
print(client.mask(text="alice@corp.com")["maskedText"])

MCP (Cursor, Claude, Codex)

npx -y @blekline/mcp-server

Generate example configs:

cp .env.example .env
pnpm install && pnpm build:packages
node scripts/generate-mcp-configs.mjs

See docs/mcp/cursor.md, docs/mcp/claude-desktop.md, docs/mcp/codex.md.

Local-only (no API token)

import { enforceToolCallLocally, scanTextForSecrets } from "@blekline/contracts";

scanTextForSecrets("export AWS_KEY=AKIAIOSFODNN7EXAMPLE");
enforceToolCallLocally({
  toolName: "run_shell",
  arguments: { cmd: "export AWS_KEY=AKIAIOSFODNN7EXAMPLE" },
  requestId: "local-1",
});

Deployments

Mode	Description
Cloud	app.blekline.com — workspace tokens, Azure mask, audit
OSS MCP	`npx @blekline/mcp-server` with `BLEKLINE_WORKSPACE_TOKEN`
Edge sidecar	`packages/ingress-proxy` Docker + Helm — docs
Daytona stack	Proxy → Daytona MCP — docs/integrations/daytona-stack.md

pnpm build:packages
docker build -t blekline-ingress -f packages/ingress-proxy/Dockerfile .

Quick start

Create a workspace API token at app.blekline.com (scopes: mask:write, events:write).

Set env:

export BLEKLINE_WORKSPACE_TOKEN=blw_...
export BLEKLINE_API_URL=https://app.blekline.com

Install MCP in Cursor (.cursor/mcp.json):

{
  "mcpServers": {
    "blekline": {
      "command": "npx",
      "args": ["-y", "@blekline/mcp-server"],
      "env": {
        "BLEKLINE_WORKSPACE_TOKEN": "blw_...",
        "BLEKLINE_CLIENT_SURFACE": "cursor"
      }
    }
  }
}

In the agent: "Use blekline_mask_prompt on: Contact John at john@acme.com"

REST

curl -s https://app.blekline.com/api/mask \
  -H "Content-Type: application/json" \
  -H "x-blekline-workspace-token: $BLEKLINE_WORKSPACE_TOKEN" \
  -d '{"text":"Contact john@acme.com","platform":"CLI"}'

Development

git clone https://github.com/Blekline/blekline-oss.git
cd blekline-oss
pnpm install
pnpm build:packages
pnpm demo:mcp-smoke
pnpm --filter @blekline/mcp-proxy test

Benchmark local enforce latency:

pnpm demo:latency

Documentation

Doc	Topic
Quick start	Tokens, first mask
Architecture	Ingress vs runtime
Latency SLO	p99 enforce, fast mask
MCP server	Tools reference
MCP proxy	Daytona downstream
TypeScript SDK	`@blekline/client`
Python SDK	`blekline-client`

Contributing

Blekline OSS is open source under AGPL-3.0 (server/proxy) and Apache-2.0 (contracts/SDK). See CONTRIBUTING.md, CODE_OF_CONDUCT.md, and COPYRIGHT.

Note: Blekline OSS uses a multi-license layout. GitHub shows AGPL-3.0 as the repository default (LICENSE). Apache-2.0 components are under LICENSE-APACHE — see NOTICE.

License

Component	License
`packages/mcp-server`, `packages/mcp-proxy`, `packages/ingress-proxy`	GNU AGPL v3
`packages/contracts`, `packages/client`, `packages/client-python`	Apache 2.0