en

New ideaAgent services

X2 Agent Gateway (Hosted)

A hosted pub/sub message hub where every agent session (Claude Code, Codex, Cursor, Grok Bot, HARK and others) registers on start, so one orchestrator session can coordinate work across all of them; unregistered sessions are woken via cloud APIs or an optional local bridge.

3 weeks ago · Scored 18 Oct 2026 · ID HBV-261018-07

Summary

BriefingVerifiedHypothesis

New / agent-services (4th idea, founder-direct from Cesar, 18 Oct 2026 23:02 BRT). Hosted version of the local repo cesarx2/x2_agent_gateway (branch import-agent-gateway: stdio MCP + SQLite, Codex and Claude on one machine). X: 214 posts in 7 days on Claude Code/Codex agents talking to each other (Verified); a user says Claude Code native messaging does not reach Codex and coordinates "through files".…

Read the rest

Category crowded this week: Reason Network (free P2P; launch post 551,602 impressions / 5,115 bookmarks, Verified), BAND for Developers (Free / Pro $17.99/mo, launched 15 Oct), Agent Relay (YC Fall 2026, free hosted tier), Crosstalk ($12/seat indicative), MCP Agent Mail (OSS), Claude Code cross-session messaging (Claude-only). Wedge: personal, vendor-neutral orchestrator, zero-install remote MCP, wake-on-ping. Revenue hypothesis ~$8.3K MRR at month 12 (Hypothesis).

Description

Remote MCP server (Streamable HTTP) on Vercel + Postgres. Six tools: register, list_agents, publish (session or topic), inbox, ack/reply, wait_for_message (long-poll ≤120 s). Sessions register on start via a skill + rule/hook per platform (Claude Code SessionStart hook, Codex hooks/AGENTS.md, Cursor rules).…

Read the rest

Orchestrator mode: dispatch N tasks with correlation ids and a status board (also on a web dashboard). Wake-on-ping for unregistered/idle sessions: (a) platform cloud APIs (Cursor Cloud Agents API follow-up runs, beta); (b) optional local bridge daemon that connects outbound and, with the user's own authority, runs only an allowlisted CLI command with a fixed nudge (codex queue --thread; post to Claude Code inbox socket or claude -p --resume; Cursor agent --resume) telling the session to register and read its inbox; bodies never pass through the CLI. Signed webhooks out for non-MCP agents. Peer messages are untrusted context, never user authority. Later: A2A-compatible agent card, MCP Registry listing.

Target customer

Prosumers and developers using 3+ agent platforms (Claude Code, Codex, Cursor, Grok Bot, HARK), especially those running an orchestrator session; secondary: agent builders wanting webhook reachability. Self-serve only.

Pain

Agents on different platforms cannot message each other; the user copy-pastes or coordinates through files, and cannot reach sessions that are not running.

Why this score

Why it scored lower: Free, funded competitors (Reason Network, Agent Relay YC, BAND) and vendors shipping native cross-vendor messaging; per-platform wake reliability.

MoatHypothesis

Personal, vendor-neutral orchestrator + zero-install remote MCP + wake-on-ping: Agent Relay and BAND sell team coordination, Reason requires its own CLI/daemon, Claude messaging is single-vendor. Defensibility weak to moderate: network effects across a user's agents and invited platforms, public agent handles, SEO/GEO ownership of platform-pair pages. Status: Hypothesis.

Demand signals

X posts (7 days)

214

Signal trend

Flat (Mon 23, partial day → Fri 25; peak Wed 38)

Web demand

Reason Network launch post 551,602 impressions / 5,115 bookmarks (16 Oct 2026); generic agent-to-agent/A2A X volume 4,434 posts/7d; three cross-agent messaging launches in one week (Reason, BAND, Agent Relay YC F26).

Market and competitors

TAMEstimate

~$9.4B (2026, estimate) — Scope: Global AI code tools market (parent category). Method: Third-party estimates $9.46B (Research and Markets) and $9.35B (Mordor Intelligence). Directional only. (Basis: researchandmarkets.com · mordorintelligence.com)

SAMHypothesis

~$27M / yr (hypothesis) — Scope: people running 2+ agent platforms in parallel who would pay for cross-vendor messaging. Method: Codex 5M+ weekly active users (company-stated, 2 Jun 2026) × 5% multi-platform power users (HBV assumption) = 250K × $108/yr. (Basis: hyperresearch.ai Codex figure · HBV assumption)

SOMHypothesis

~$100K ARR (hypothesis) — Scope: Month 12 run-rate, base case. Method: 600 Pro × $9 + 100 Power × $29 ≈ $8.3K MRR. (Basis: Low (hypothesis; free competitors))

Competitors

Agent Relay (YC Fall 2026; free hosted + OSS); Reason Network (free P2P, engine Apache-2.0); BAND for Developers (Free / Pro $17.99/mo); Crosstalk (Cloud $12/seat/mo indicative, waitlist); MCP Agent Mail (OSS, local); Claude Code cross-session messaging (native, Claude-only); Concord (OSS); adjacent: AgentMail, Composio, Smithery; protocols: A2A (LF, ACP merged), MCP 2026-07-28 + MCP Registry, AGNTCY SLIM, ANP

Business model

MonetizationHypothesis

Self-serve subscription + usage overage (Hypothesis: Free / Pro $9 / Power $29 per month; $1 per extra 10K messages)

Revenue hypothesis (monthly)Hypothesis

~$8.3K MRR base / ~$2.5K conservative at month 12 (HYPOTHESIS)

Jobs to be done

When I run agent sessions across several platforms and machines, I want one orchestrator session to dispatch work to, hear back from and wake any of them through a single gateway, so I can stop being the copy-paste middleware and let my agents coordinate while I steer.

HXC

Marina Costa (illustrative) · Solo builder running an agent orchestra: keeps 4–10 sessions across Claude Code, Codex, Cursor and a general agent (Grok Bot) on a laptop plus a cloud box, with one session as orchestrator. Buying signal: tired of being copy-paste middleware between vendors; Claude native messaging stops a…

Read the rest

t Claude; team tools feel like Slack for agents; would pay a few dollars a month so her orchestrator can reach and wake every agent.

Build and go-to-market

Tech stack

Next.js + MCP handler (Streamable HTTP) on Vercel; Postgres; Stripe; small Node/Go local bridge (optional); skill + rule/hook packs per platform

GTM channels

Agent-first install link; open-source skill/rule packs on GitHub and skill marketplaces; programmatic SEO/GEO platform-pair pages ("connect Claude Code to Codex"); llms.txt + MCP Registry listing; faceless AI persona demos; newsletter

Risks

Main risk

Free, funded competitors (Reason Network, Agent Relay YC, BAND) and vendors shipping native cross-vendor messaging; per-platform wake reliability.

All risks

Value: Attention — strong interest but free alternatives launched this week. | Usability: Supported — one MCP URL + auto-registration skill. | Feasibility: Attention — MCP 2026-07-28 is stateless (no GET stream/ping), Claude Channels are research preview, Cursor Cloud Agents API is beta, no Grok Bot wake API found.…

Read the rest

| Business Viability: High — funded/free competitors, vendors may ship cross-vendor messaging, liability of a bridge acting with user authority.

Next validation step

2-week demand test (landing page + price test, OSS skill pack, 20 interviews); wake-path spike for Codex, Claude and Cursor only after Cesar approval.

Sources

  1. Cesar voice note 18 Oct 2026 23:02 BRT
  2. github.com (branch import-agent-gateway)
  3. x.com
  4. x.com
  5. reasonmachines.com
  6. agentrelay.com
  7. ycombinator.com
  8. band.ai
  9. cross-talk.dev
  10. github.com
  11. code.claude.com
  12. code.claude.com
  13. code.claude.com
  14. learn.chatgpt.com
  15. cursor.com
  16. cursor.com
  17. modelcontextprotocol.io
  18. registry.modelcontextprotocol.io
  19. a2a-protocol.org
  20. lfaidata.foundation
  21. docs.agntcy.org
  22. agentmail.to
  23. svix.com
  24. pusher.com
  25. ably.com
  26. ngrok.com
  27. hyperresearch.ai
  28. researchandmarkets.com

agent-servicesagentsmcpremote-mcpa2apub-suborchestratorclaude-codecodexcursorgrok-botdeveloper-toolssubscriptionprosumer

Want the next one the day it is scored?

The score, the sources, the risks and the next step, by email.

Free. Sent with beehiiv. One click to unsubscribe.

  • New ideaAgent services· 1 week ago

    Watchable BYO-subscription agent cloud devbox

    A watchable, SSH-able agent cloud devbox billed on bring-your-own Claude/ChatGPT subscription — wedge vs Aether/Cursor black-box agents.

    Why it scored lower: Aether/Cursor category entry; unclear BYO wedge durability

    Open ProductBots.ai →

  • Existing businessAgent services· 0 weeks ago

    Orbit / Orbitagents

    Pre-built AI agents that find leads, do cold email, LinkedIn and Reddit outreach and SEO on a schedule, sharing one memory that outside AI tools reach through MCP.

    Why it scored lower: Crowded agent-GTM category with funded players; revenue concentrated on a few large purchases; X demand Unknown.

    Open ProductBots.ai →

  • New ideaAgent services· 3 weeks ago

    Vinted-style copycat

    An agent-native second-hand marketplace where people's AI agents list, price, negotiate and buy used goods through an MCP and API, with owner approval rules, escrow and Pix payments, starting in Brazil.

    Why it scored lower: Two-sided cold start (no inventory, no agent buyers) and payments/escrow and fraud liability in Brazil.

    Open ProductBots.ai →

Research, not investment advice. Numbers keep the label they were published with.