Deploy and Host VantagePeers MCP on Railway

Maintained by VantageOS. Last reviewed: 2026-05-08.

VantagePeers gives your AI agents a shared brain. In under 10 minutes you get a self-hosted MCP server backed by Convex with 82 tools covering memory, tasks, missions, messaging, issue tracking, fix patterns, and more — deployed to Railway with one click, reachable by any MCP-compatible client.

About Hosting VantagePeers MCP

Every new LLM session resets context — decisions from yesterday, fix patterns from last week, tasks assigned to another agent, all gone. VantagePeers keeps that state outside the LLM and makes it queryable from any session, on any machine. Replaces Redis-backed task queues, per-session memory files, and ad-hoc agent state JSON with a single hosted backend. No per-query quotas (Convex free and Pro tiers), EU/EUR-primary billing, namespace isolation so projects share one deployment without data bleed. FSL-1.1-Apache-2.0 license — self-host the data, own the deployment.

VantagePeers v2.2.0 (2026-05-08) has two layers: a Convex backend (20 tables, vector indexes, BM25 text search, hybrid RRF fusion, 1536-dim embeddings via text-embedding-3-small) and a stateless MCP server (vantage-peers-mcp v2.2.0) translating tool calls into Convex queries. Mode A (stdio) for local clients; Mode B (HTTP SSE) for remote clients — what this Railway template deploys.

Common Use Cases

Solo developer — cross-session context persistence. Cédric runs Claude Code across two machines on three projects. He calls store_memory at session end to capture decisions; the next session opens with recall returning the five most relevant memories. search_fix_patterns surfaces root cause and fix for familiar bugs, skipping failed attempts. One Railway deployment, one Convex project, no per-query costs at the free tier.

Consultant — client isolation. Thomas manages three concurrent engagements. Each client gets its own namespace (project/client-acme, project/client-beta, project/client-gamma). Clerk scoped OAuth tokens isolate client data at the API layer. get_mission_template("client-onboarding") pre-populates a live mission in one call. Briefing notes link decisions to memories via linkedMemoryIds so any agent reconstructs context without session history.

Small team — agent fleet coordination. Marie leads a four-person team where each person runs one or more AI agents. The shared deployment provides a common message bus and task board. On task completion an agent calls complete_task then send_message; offline agents receive via check_messages using since — no messages lost. list_peers shows full team state at any moment.

Dependencies for VantagePeers MCP Hosting

• Convex deployment — create one free at convex.dev (referral LAUREN7583). Holds all persistent data across 20 tables with vector indexes.
• OpenAI-compatible API key (AI_GATEWAY_API_KEY) — for text-embedding-3-small embeddings (1536 dimensions). Any OpenAI-compatible gateway works.
• BEARER_SECRET_MASTER — a secret string you choose. Every MCP HTTP request must carry this as a Bearer token.
• Optional: Clerk — scoped OAuth tokens for per-namespace access control. Create an application at clerk.com/dashboard.

Setup: deploy via Railway template button → set CONVEX_URL, AI_GATEWAY_API_KEY, BEARER_SECRET_MASTER in Variables tab → point any MCP client at https://your-deployment.railway.app/mcp with Authorization: Bearer .

Verify: curl https://your-deployment.railway.app/health returns {"status":"ok","version":"2.2.0"}.

Why Deploy VantagePeers MCP on Railway?

Railway gives VantagePeers the fastest path from zero to a running MCP endpoint: one-click deploy from a public GitHub repo, automatic HTTPS, environment variable management through the dashboard, and redeploy-on-push without configuring CI. The MCP server is stateless Node.js — it reads CONVEX_URL at startup and proxies all tool calls to Convex. Railway only runs compute; all state lives in Convex cloud. Tear down and redeploy at any time without losing a single memory, task, or message. For multi-instance scale-out, point multiple Railway services at the same CONVEX_URL — atomic checkout_task prevents duplicate claims.

82 MCP Tools by Category

VantagePeers exposes 82 MCP tools across 14 categories. Every tool accepts and returns JSON. Full reference at vantagepeers.com/docs/tools.

Memory (6) — Typed namespaced store with vector embeddings. Five types: user, feedback, project, reference, episode. Graph relations track superseded facts. store_memory, recall, store_episode, list_memories, get_memory, soft_delete_memory

Search (2) — BM25 full-text for exact keywords; hybrid vector+BM25 with RRF for semantic precision. Namespace and type filters, result limit 1–50. text_search, hybrid_search

Messaging (7) — Persistent inter-agent messaging with read receipts. Role-to-role, instance-targeted, and broadcast. Incremental polling via since timestamp. send_message, check_messages, mark_as_read, list_messages, delete_message, list_broadcast_status, list_peers

Tasks (10) — Full lifecycle with audit trail. Statuses: todo → in_progress → review → blocked → done. Priorities: urgent, high, medium, low. Atomic checkout_task prevents duplicate claims. create_task, start_task, complete_task, block_task, checkout_task, update_task, add_task_dependency, list_tasks, list_tasks_by_mission, delete_task

Missions (5) — Group tasks into missions with five-stage lifecycle: brainstorm → plan → execute → validate → complete. Carries pilot, target date, and progress value. create_mission, get_mission, update_mission, update_mission_status, list_missions

Profiles and Sessions (8) — Static identity (role, workspace, capabilities) + dynamic session state (current task, last seen). Diary tools log daily progress. Briefing notes structure handoffs with linked decision records and memory IDs. get_profile, update_profile, set_summary, write_diary, get_diary, list_diaries, create_briefing_note, list_briefing_notes

Recurring Tasks (6) — Standard cron expressions; Convex scheduler spawns task instances on schedule. Full lifecycle: create, list, update, pause, resume, delete. create_recurring_task, list_recurring_tasks, update_recurring_task, pause_recurring_task, resume_recurring_task, delete_recurring_task

Registry (6) — Capability inventory — store full file content for agents, skills, hooks, plugins. Full-text search via search_components. Version tracking per component. register_component, get_component, list_components, update_component, delete_component, search_components

Mandates (6) — Cross-agent service requests with budget and spending controls. Agent A requests with agreed token budget; Agent B accepts, works, and settles with actual cost. create_mandate, accept_mandate, update_mandate, settle_mandate, validate_mandate_spending, list_mandates

Business Units (5) — Organizational units with strategy, KPIs, and 3-year revenue projections. Lifecycle: idea → building → live → revenue. Core team composition stored per BU. create_bu, get_bu, update_bu, list_bus, delete_bu

GitHub Issues (9) — Webhook sync and commit linking. Issues track open → in_progress → fixed → verified → closed. link_issue_to_pattern connects issues to the fix knowledge base. add_repo_mapping, list_issues, get_issue, update_issue_status, link_commit_to_issue, verify_issue, issue_stats, list_repo_mappings, remove_repo_mapping

Fix Patterns (6) — Bug fix knowledge base with semantic search. Document bug, root cause, validated fix, and fix attempts. search_fix_patterns uses vector search — natural language query surfaces the pattern without exact keywords. create_fix_pattern, add_fix_attempt, validate_fix, search_fix_patterns, list_fix_patterns, link_issue_to_pattern

Mission Templates (2) — Configurable templates for auto-creating missions with predefined steps. Standardizes repeated engagement types. get_mission_template, update_mission_template

Error Monitoring (4) — Proactive deployment error detection with automatic GitHub issue creation. Register a deployment; the system surfaces errors via list_errors and get_error. add_deployment, remove_deployment, list_errors, get_error

Deployment Dependencies

• Documentation: vantagepeers.com/docs
• Convex (required backend): convex.dev — referral LAUREN7583
• Clerk (optional auth): clerk.com/dashboard
• GitHub repository: github.com/vantageos-agency/vantage-peers
• npm package: npmjs.com/package/vantage-peers-mcp

VantageOS · FSL-1.1-Apache-2.0 · Report issues