Exa.ai — the real deal

In a sea of MCP hop-on bloat — “connected!”, forty tools, zero rent — Exa is the first research product we will actually call extremely good. Semantic search that finds the right page. Clean extracts. Not SEO soup. Not vendor theater.

Titanium CLIs: no MCP policy. Codex Titanium / sekhmet / L3 sparks do not take an MCP zoo. This page is not an instruction to wire Exa (or any MCP) into Titanium. Keep the substrate pure: namespaced sparks, Rust, no worktree theater, no MCP bloat on the L3 host.

The real deal: Exa is legitimately excellent for web research — product docs, papers, eligibility, competitive maps, “what does this page actually say.” First thing in the MCP hype cycle that felt like a sharp tool instead of a plugin flea market. If you use research hosts that already speak Exa’s API/tools, great. If you run Titanium — stay no-MCP; use Exa outside that path when you need the web brain.

Extremely good at

Research quality

Semantic hit rate — describes the page you want, not keyword bingo
Clean fetch — readable extract, not a DOM landfill
Primary sources — good at landing docs, papers, official pages
Operator speed — fewer retries than laggy “web MCP” clones

What we refuse to do

Ship MCP into Titanium — against DS4CC L3 policy
Enable a zoo — twelve half-dead servers ≠ intelligence
Confuse “connected” with “useful” — auth green lights mean nothing
Replace live UI — burner browser still wins for real buttons

Where Exa belongs

Human + research agents on hosts that already integrate Exa
Operator sessions that need the open web, fast
Alongside — not instead of — sekhmet/Titanium for pure exec

Where it does not

Codex Titanium / sekhmet L3 worker policy (no MCP)
“Default stack = more MCPs” marketing
Replacing local authority (SQLite, git, tests) with cloud memory theater

Product: exa.ai · Marketplace home: ds4cc.com · Anti-patterns: ANTI-PATTERNS.md