Hyperengage ships an MCP server that exposes live customer state as typed tools. Claude, Codex, ChatGPT, or your own agent can read events, emails, meetings, decisions, actions, and threads. And write back. No stale CRM exports. No glue code.
{
"mcpServers": {
"hyperengage-memory": {
"command": "npx",
"args": ["@hyperengage/mcp"],
"env": {
"HYPERENGAGE_API_KEY": "hp_live_••••••••"
}
}
}
}Why MCP
A CRM export is a snapshot. By the time your agent reads it, the customer has shipped a feature, opened a ticket, sent a renewal email, and changed champion. Your agent acts on a world that no longer exists.
The Customer Memory Graph reconciles every signal as it happens. MCP gives your agent a tool to read it the moment it's true. Live state. Reasoned. Tenant-scoped. Always current.
CRM export
Memory Graph
How it works
STEP 01
Add the Hyperengage MCP server to your agent's mcp.json. Two lines for Claude, Codex, Cursor, or Continue. Done in 60 seconds.
STEP 02
memory.read, memory.search, memory.subscribe, graph.traverse, memory.write — auto-discovered, fully typed, with example usage.
STEP 03
The agent calls tools as needed. Reads return reasoned, reconciled state. Writes update the graph. Sidekick, Copilot, and Signals Engine see the changes instantly.
What your agent reads
The graph stores six entity types per customer. Your agent reads them as structured, typed objects with relationships, sentiment, and derived signals.
Events
Product usage, feature adoption, milestone hits.
Emails
Inbound and outbound across every CSM and customer contact.
Meetings
Recordings, transcripts, and reasoned briefs — not raw notes.
Decisions
Renewal calls, pricing concessions, exec asks. Captured and timestamped.
Actions
Playbook steps, follow-ups, CRM updates. What was done, by whom, when.
Threads
Slack and internal conversations about the account, threaded to it.
Tools your agent gets
memory.read
Pull a customer's full memory or filter by entity type, time window, or signal.
memory.search
Semantic search across every customer interaction — emails, meeting transcripts, threads.
memory.subscribe
Stream changes as they reconcile. Your agent reacts in seconds, not on a polling loop.
memory.write
Write back decisions, actions, and notes. The graph is bidirectional.
graph.traverse
Walk relationships — account → contacts → meetings → decisions → next actions.
graph.diff
Ask what changed between two points in time. Reasoned diffs, not raw deltas.
Built right
Live state, not stale exports
Your agent sees the graph as it is right now — not a CSV exported six hours ago.
Bring any agent
Claude, Codex, ChatGPT, or your own custom agent. MCP is the universal protocol.
Reasoned, not raw
Tools return interpreted state — sentiment, trajectory, signals — not just rows.
Tenant-scoped & audited
Every call is workspace-scoped, logged, and rate-limited. Read scopes are explicit.
Bidirectional
Agents read state and write decisions/actions back. The graph stays in sync with what your agents do.
Drop-in MCP
Stand up an MCP server in two lines of config. Works with any MCP-compatible host.
Works with
If it speaks MCP, it can read the graph. We ship example configs for the runtimes our customers actually use.
Claude
Claude Code
Codex
ChatGPT
Cursor
Custom
FAQ
Model Context Protocol — an open standard for letting AI agents call typed tools across servers. Hyperengage exposes the Customer Memory Graph as an MCP server, so any MCP-compatible agent (Claude, Codex, ChatGPT desktop, your own runtime) can read live customer state.
A CRM API returns whatever someone typed into a form. The Customer Memory Graph returns reconciled, reasoned state — sentiment, trajectory, risk signals, expansion moments — fused across CRM, product, support, and comms. Your agent doesn't have to stitch context together itself.
Yes. memory.write lets agents persist decisions, actions, follow-ups, and notes. The graph stays consistent with what your agents do, and Sidekick / Copilot / Signals Engine see the same writes immediately.
Anything that speaks MCP. Confirmed: Claude (claude.ai, Claude Code), Codex, ChatGPT (desktop), Cursor, Continue, and custom runtimes built on the Anthropic / OpenAI / Vercel AI SDKs. We ship example configs for each.
MCP access is included with every Hyperengage workspace. Per-call rate limits scale with your plan. Beta participants get unlimited calls during the program.
MCP is an open standard. Our implementation follows the spec; tool schemas, scopes, and webhooks are documented. You're never locked into Hyperengage's stack to talk to the graph.
We onboard a small batch each month. Tell us what you're building and we'll fit you in.