Agent Prompt Injection Firewall Mcp
Scan inbound prompts, RAG chunks, and tool arguments with a WAF-style firewall before your agent acts on untrusted text.
Overview
Agent Prompt Injection Firewall MCP is a MCP server for the Ship phase that scans prompts, RAG content, and tool arguments to block prompt-injection attacks on agents.
What is this MCP server?
- WAF-style firewall for agents: pattern-based plus heuristic scanning
- Inspects prompts, RAG documents, and tool arguments on the MCP path
- Streamable HTTP remote and stdio PyPI package at v1.0.5
- Python stdio transport for local MCP clients
- Hosted endpoint: api.meok.ai v1/a2a/prompt-injection-firewall
- Server version 1.0.5
- 1 streamable-http remote
- 1 PyPI stdio package with Python runtimeHint
What problem does it solve?
Untrusted text in prompts or retrieval can smuggle instructions that hijack your agent’s tool use without touching your application code.
Who is it for?
Builders launching RAG-heavy or user-facing agents who want a dedicated injection filter at the MCP layer.
Skip if: Fully offline, trusted-input-only dev prompts where you never ingest external documents or end-user text.
What do I get? / Deliverables
After registration, suspicious patterns and heuristics are evaluated on the MCP path so risky content can be stopped before tools execute.
- MCP-accessible injection scanning on agent-bound content
- Defense layer for prompts, documents, and tool parameters
- Hosted or local firewall endpoint in your agent toolchain
Recommended MCP Servers
Journey fit
Prompt-injection defense is filed under Ship → Security because it is a production control you add before exposing agents to user or document-sourced input. Security subphase reflects pattern-based and heuristic blocking of malicious instructions, not general LLM feature work.
How it compares
Agent-input WAF MCP, not a code-review skill or generic moderation API wrapper alone.
Common Questions / FAQ
Who is agent-prompt-injection-firewall-mcp for?
Developers and indie operators running MCP-connected agents that consume user input, web content, or RAG documents and need injection screening.
When should I use agent-prompt-injection-firewall-mcp?
Use it as soon as external or retrieved text can reach your agent’s context or tool arguments—before public launch and continuously in production.
How do I add agent-prompt-injection-firewall-mcp to my agent?
Install the PyPI package agent-prompt-injection-firewall-mcp for stdio MCP, or configure streamable-http to https://api.meok.ai/v1/a2a/prompt-injection-firewall.