What changed
Remote MCP is becoming a managed, provider-supported primitive (Gemini API now natively manages remote MCP connections and background agents — FACT from blog.google source), and Anthropic-API-compatible backends (Ollama's Claude Code compatibility — FACT from ollama.com source) mean agent setups are standardized enough that a memory server can drop in without custom integration work.
Why now
Claude Code team usage is growing faster than its collaboration primitives. Context accumulates per-machine per-developer with no sanctioned sync mechanism, and the MCP protocol now gives a standard socket to plug a memory layer into. The window is before Anthropic ships native team memory.
Converging signals
(1) Concrete PAIN: HN post describing two cofounders with drifting per-machine Claude Code 'company brains', no shared/private boundary, hacky Google Drive workaround (FACT, cited). (2) Remote MCP as managed primitive lowers solo shipping cost (FACT, headline-level). (3) Anthropic-API-compatible local backends broaden the addressable install base (FACT).
Customer pain
Teams of 2-10 running Claude Code each accumulate strategy, customer notes, decisions, and conventions inside one machine's context. Teammates re-derive answers from zero, the two brains drift, and there is no history, no merge/conflict handling, and no way to share a subset while keeping personal context private. This is verbatim from the PAIN source, not inferred.
Who pays
HYPOTHESIS: AI-native startup founders and small dev teams (2-10 seats) who already pay for Claude Max/API and treat agent context as a company asset. The single PAIN source proves the problem exists for at least this profile; it does not prove willingness to pay.
Solved today
Per the PAIN source: hacky workarounds (a shared Google Drive folder). More broadly (HYPOTHESIS from ecosystem knowledge): git-committed CLAUDE.md/memory files, copy-pasted context, or wiring up memory frameworks like mem0/Zep/Letta themselves.
Why current solutions are bad
File-sync approaches have no shared/private scoping, no conflict resolution, no history, and no semantic retrieval; DIY memory-framework wiring is real engineering work each team must repeat and none of it is Claude-Code-native out of the box.
Proposed product
A hosted (and self-hostable) MCP memory server: teams connect Claude Code via one MCP config line; memories are written/read through MCP tools with 'shared' vs 'private' scopes, append-only history with rollback, embedding-based retrieval, and last-writer-wins-plus-audit conflict handling. Per-team pricing, SOC2-later.
MVP version
Single-tenant remote MCP server (FastAPI + Postgres + pgvector — the founder's existing stack) exposing store/recall/search tools with scope tags and a history log, plus a one-page web viewer. No dashboard polish, no SSO. 2-4 weeks of AI-assisted build.
30-day build
Ship MVP; recruit 10 design partners directly from the HN thread and Claude Code Discord/Reddit communities; instrument what memories teams actually share vs keep private.
60-day build
Add conflict/merge UX, team invites, and export/portability (the anti-lock-in wedge vs mem0/Zep); publish to MCP server directories; Show HN launch.
90-day revenue plan
Convert design partners to $29-49/team/month; realistic first revenue day 60-120. HYPOTHESIS: 10-30 paying teams by day 180 (~$500-1,500 MRR) — modest, and that is the strategic problem.
Distribution path
Show HN (the pain post proves the audience is there), MCP server directories/registries, Claude Code subreddit/Discord, comparison content vs mem0 OpenMemory. No paid acquisition needed.
Pricing hypothesis
$29-49 per team/month hosted; one-time $499 self-hosted license as a secondary SKU for privacy-sensitive teams. Per-seat expansion above 5 users.
Technical difficulty
Low-moderate for this founder: MCP server + Postgres/pgvector + auth is squarely within his demonstrated FastAPI/Postgres/agent-tooling capability (he built this very engine on that stack).
Legal / regulatory risk
Low. Storing customers' business context creates ordinary data-protection obligations (DPA, encryption), no regulated data by default.
Platform dependency
HIGH and this is the core risk: the product lives inside Anthropic's ecosystem. Anthropic already ships memory features and could release native team/shared memory at any model or Claude Code update, instantly commoditizing this. MCP portability across providers (Gemini remote MCP support — cited) partially hedges but does not remove it.
Founder fit
Mismatched with his proven edge. His demonstrated wedge is forced-buyer government-portal filing automation (FMCSA ELDT) — structural demand with deadlines. This is a discretionary dev-tool sold to developers, the buyer class most prone to DIY (the pain itself is solvable with a git repo of markdown files), in a space where his industrial/public-records/compliance credibility earns nothing. The 0.8-confidence lesson that gov-portal mandate opportunities score 8-9 founder-fit applies here in the negative: this is not that shape. He CAN build it; it is not what he is uniquely positioned to sell.
Breakout potential
If it works, expansion into cross-provider agent memory (Claude+Gemini+local models) is real, and 'memory layer for agent teams' is a venture-scale narrative — which itself is a warning sign: venture-funded competitors (mem0 et al.) are already there.
Final recommendation
PASS for this founder — not for lack of pain, but for lack of a defensible wedge he specifically can hold. The demand evidence is one complaint; existing spend is unproven; incumbents and the platform owner both loom; and it spends his runway on a crowded discretionary dev-tool instead of his proven forced-buyer government-filing lane. Revisit only if 60 days from now Anthropic has NOT shipped team memory and the HN thread has spawned visible paying alternatives (which would prove the market).
Next action
Spend one hour, not one sprint: reply to the HN thread asking the OP and commenters 'would you pay $29/mo for X', and check mem0/Zep positioning. If ≥5 independent teams say yes-with-credit-card, upgrade to revisit; otherwise archive and keep the reasoning budget on mandate-shaped opportunities.