Why do models sometimes omit headers like Content‑Length or URL encoding?
This is a known model limitation, not a bug in BurpMCP. The LLM occasionally fails to add required components to requests.
@swgee
基本情報
Burp Suite Extension with MCP Server to enhance manual application security testing
以下の設定を使って、このサーバーを MCP 対応クライアントに追加してください。
{
"mcpServers": {
"burpmcp": {
"autoApprove": [],
"disabled": false,
"timeout": 30,
"url": "http://localhost:8181/mcp/sse",
"transportType": "sse"
}
}
}ツール一覧はまだありません
この掲載には表示できるツール一覧がまだありません。サーバーにツールがないという意味ではありません。対応機能はプロジェクトのドキュメントをご確認ください。
上のボタンでサーバーのエンドポイントに接続し、ツール一覧を取得できます。
BurpMCP is a Burp Suite extension that integrates with Model Context Protocol (MCP) clients like Claude Desktop and Cursor, allowing LLMs to assist in manual HTTP-based application security testing. It provides tools to save, send, and modify requests, generate Collaborator payloads, and view debug logs, serving as an AI-powered sidekick for security testers, vulnerability researchers, and bug bounty hunters.
Download the jar from releases and load it into Burp. The MCP server runs by default on localhost port 8181 over SSE. Configure your MCP client (e.g., Cline, Dive) with the SSE URL http://localhost:8181/mcp/sse. For STDIO‑only clients like Claude Desktop, use the provided stdio-bridge.py script. To send a request, right‑click it in Burp and select Extensions -> Send to BurpMCP. Saved requests appear in the “Saved Requests” tab and can be retrieved by the LLM.
This is a known model limitation, not a bug in BurpMCP. The LLM occasionally fails to add required components to requests.
Examine the request for forbidden headers. The HTTP/2 tool parameter specifies that such headers should be excluded, but the model may ignore this instruction.
Some MCP clients cannot send `\r\n`. BurpMCP can automatically replace LF with CRLF. Enable this feature in the extension settings.
Some MCP clients do not acknowledge error responses promptly. The best workaround is to use a client that supports timeouts, such as Cline.
Headers are joined by newlines and re‑split before sending, and cookies become separate headers. This makes manual injection of newlines into header values difficult, limiting HTTP/2 protocol vulnerability testing.

SSH MCP server for AI agents: remote commands, file transfer, log search and server audits through OpenSSH.

Open-source Next.js + Supabase CMS with a per-site MCP server: scaffold a site, then create pages, posts, navigation, media and SEO from Claude, Cursor or any MCP client. Streamable HTTP with a bearer token you mint in y

Automated codebase documentation for GitHub: searchable docs, MCP context for AI agents, doc drift detection, and Friday Cleanup PRs.

Real UI references for AI coding agents: 500,000+ real screens, 15k+ website design systems with a DESIGN.md, and 30k+ UI elements, so your agent builds UI that looks real, not AI slop.
Agent conductor — AGENTS.md contract parser, SKILL.md registry, CHP decision engine over MCP
コメント