MCP File Upload: Why Base64 Tool Arguments Can Corrupt Files
A 3 KB PNG lost 32 characters on the way into a tool call. The server said success.
October 03, 2026 · 9 min
Latest · Post
PDF Prompt Injection: Hidden Text Only Your AI Agent Sees
White 0.1pt text told AI reviewers what to write. Geometry catches what classifiers miss.
Latest — recent dispatches
View all →
Extract Chart Data From PDFs: An MCP Tool That Won't Guess
Exact numbers come from the server's geometry. The judgment calls get handed back to the LLM.
September 26, 2026 · 11 min
Cut LLM Judge Cost Without Adding Bias
Two shortcuts cut my judge calls. One kept the verdicts. The other quietly flattered them.
September 19, 2026 · 12 min
Most read guides — reader favorites
View archive →
A Knowledge Base Is Just a Folder
No pipeline, no vector store. Topic folders, a warm habit, a generated catalog, honest coverage.
August 01, 2026 · 10 min
How AI Agents Should Read PDFs: 5 Patterns That Survived Production
How agents should navigate documents at production scale, from 50,000+ downloads of one MCP server.
June 24, 2026 · 11 min
How to Ship an MCP Server to Production (Authentication, Safety, and What Breaks)
MCP tutorials stop at tool integration. The real work is auth, guardrails, and untrusted data.
April 11, 2026 · 7 min
How Claude Code Actually Reads PDFs: Lessons from Building an MCP Server
Tool boundaries shape agent behavior more than prompts. What I learned shipping pdf-mcp.
April 04, 2026 · 10 min
How to Build an MCP Server in Python with FastMCP 4 (Full Code)
Tools, resources, prompts, SQLite persistence, in-process testing. Under 100 lines of Python.
February 22, 2026 · 9 min
pdf-mcp: How to Handle Large PDFs in Claude Code with MCP
Past 100 pages, Claude reads a PDF as text only. pdf-mcp reads in chunks.
January 31, 2026 · 16 min
Parallel Processing in AWS Lambda with Python: What Actually Works (And What Doesn't)
What threading, multiprocessing, and async actually do inside a Lambda function. Benchmarked.
January 17, 2026 · 4 min
How to Connect a Legacy System to AI Agents Using MCP
How to wrap a legacy Redmine API with MCP. No rewriting, no destabilizing anything.
May 27, 2025 · 6 min
More Articles
01
How Amazon Bedrock Helped Me Make My RAG Better
02
How to Send PDFs Over 100 Pages to Claude's API
03
How to Test and Debug a Python MCP Server
04
How I Fixed Vertical Japanese PDF Extraction
05
WebMCP in Production: What I Shipped and What I Removed
Highlight Editor's pick →
How I Built a Serverless Newsletter on AWS for Under $1/Month
Seven Lambdas, two SQS queues, one DynamoDB table. SES for sending. No per-subscriber fee.
June 20, 2026 · 10 min
MCP Tool Sprawl: How I Cut 69 Tools to 43 With a Decorator
Why more MCP tools make agents worse, and the pattern that fixes the surface.
June 13, 2026 · 8 min
Section Chunking vs Page Chunking for AI Agents: ~6 Fewer Tool Calls Per PDF Query
Page-mode PDF search costs 2 to 6 extra tool calls per query depending on the document. Section-aware search delivers...
May 30, 2026 · 10 min
MCP Server Security: 8 Vulnerabilities I Found and Fixed
I audited my MCP server at 3,000 downloads and found 8 vulnerabilities. Every fix is here.
February 25, 2026 · 10 min
MCP vs Function Calling: Key Differences, Trade-offs, and When to Use Each
Function calling couples tools to one app. MCP decouples them. When each approach breaks down.
February 16, 2026 · 11 min
AI Agent API Access: Why Full Permissions Are a Security Risk
Three MCP design patterns from giving an agent full API access. And watching it go wrong.
December 24, 2025 · 4 min
I Tried AWS Bedrock Knowledge Bases: What Worked and What Didn't
20-minute setup, then an unexpected OpenSearch Serverless bill. What it actually delivers.
April 14, 2025 · 10 min
AWS Global to China S3 Sync: A Serverless Solution
Event-driven sync across the China boundary. No VPN, no persistent infrastructure.
April 12, 2025 · 5 min