Blog

Notes on agent memory, MCP, and building threadctx.

2026-07-10

Your company blocked an MCP server. Good.

Admins are right to review MCP servers — they run inside the agent loop. The four questions any MCP server should answer, and how to get a legitimate one approved.

2026-07-06

threadctx works with your custom agents too — we do the integration if you need it

threadctx isn’t limited to Claude Code, Cursor, LangGraph, or Agent Framework. Any agent that can speak MCP can use it — including whatever your team built in-house. If yours can’t yet, we’ll help.

2026-07-06

Your agents don’t have a memory problem. They have a team memory problem.

Most “agent memory” is built to personalize one agent to one user. That’s a different problem from what a dev team actually needs.

2026-07-06

What actually deserves an MCP slot in your Claude Code or Cursor setup

MCP config is a limited-attention budget, not an unlimited plugin list. Here’s how to think about what earns a slot.

2026-07-05

MCP is becoming the universal interface for agent memory

Memory is decoupling from any single agent framework the same way language tooling decoupled from any single editor — and MCP is the protocol making it happen.

2026-07-05

Give your LangGraph agents shared team memory in 5 minutes

A step-by-step guide to wiring threadctx into a LangGraph agent with langchain-mcp-adapters — no custom integration required.

2026-07-05

One memory layer across Claude Code, Cursor, and LangGraph

What it actually looks like when one engineer’s agent — in any framework — learns something and every other agent on the team can recall it.