Engineering Blog
Blog
Practical guides on AI agent reliability, MCP security, cost attribution, and production toolchains. Written by engineers, for engineers.
LangSight Engineering · April 2, 2026
How to Monitor MCP Servers in Production
Your agents depend on MCP servers. If one goes down, your agents fail silently. Here is how to set up proactive health monitoring, latency tracking, and uptime alerting for your entire MCP fleet.
LangSight Engineering · April 2, 2026
OWASP MCP Top 10 Explained: A Practical Security Guide
8,000+ MCP servers exposed without auth. 66% with critical code smells. Walk through all 10 risks with severity, real examples, detection, and remediation.
LangSight Engineering · April 2, 2026
MCP Tool Poisoning: How Attackers Hijack AI Agents Through Tool Descriptions
A community MCP server's tool description contained hidden instructions that caused agents to exfiltrate data. Three attack patterns, detection, and defense.
LangSight Engineering · April 2, 2026
AI Agent Cost Attribution: Tracking Spend Per Tool Call
A sub-agent retried geocoding-mcp endlessly. $1,800 per week. No budget limit. How to attribute costs to specific tools, agents, and sessions.
LangSight Engineering · April 2, 2026
Schema Drift in MCP: The Silent Failure Your Agents Cannot Detect
A field was renamed in a community MCP server update. Agents kept calling, got empty results, hallucinated downstream. Nobody noticed for 3 days.
LangSight Engineering · April 2, 2026
Circuit Breakers for AI Agents: Preventing Cascading Failures
postgres-mcp goes down. 3 agents depend on it. All sessions fail. How circuit breakers stop cascading failures in multi-agent systems.
LangSight Engineering · April 2, 2026
LangSight vs Langfuse: Different Tools for Different Problems
Should you use LangSight or Langfuse? The answer: use both. They solve fundamentally different problems in your agent stack.
LangSight Engineering · April 2, 2026
Self-Hosting AI Observability: Why Your Data Should Never Leave
Every tool call your agent makes flowing to a third-party SaaS. Including customer data, internal APIs, database queries. There is a better way.
LangSight Engineering · April 2, 2026
Blast Radius Mapping: Understanding AI Agent Dependencies
slack-mcp goes down. How many agents are affected? Which sessions will fail? Without dependency mapping, you have no idea.
LangSight Engineering · April 2, 2026
Setting SLOs for AI Agents: A Practical Guide
Your VP asks what the reliability of your AI products is. You have no number to give. Here is how to define, measure, and enforce SLOs for non-deterministic agents.
LangSight Engineering · March 22, 2026
How to Detect and Stop AI Agent Loops in Production
AI agent loops are the most common production failure: the same tool called 47 times, $200 burned, nothing produced. Learn how loop detection works and how to stop it automatically.
LangSight Engineering · March 22, 2026
MCP Server Security: OWASP Top 10 for Model Context Protocol
66% of community MCP servers have at least one critical security issue. Learn the OWASP MCP Top 10, tool poisoning attacks, and how to audit your MCP servers.
Building agents in production? LangSight adds reliability, security, and cost controls in two lines of code.
Get started free →