Durable evidence
Retrieve project-specific facts and provenance with relevance signals and an explicit no-evidence state.
Governed integration layer
A secure memory interface for trusted AI systems.
Eventium MCP gives approved applications a small, standards-based interface to durable Atlas memory and project rules. Identity, authorization, scope, and physical memory boundaries remain server-owned.
Seven allowlisted tools. No browser credentials, cross-user namespaces, or unrestricted tool execution.
GatewayChecking…
Atlas connectionChecking…
ProtocolMCP Streamable HTTP
Version0.2.0
Capabilities
Useful memory, bounded by policy.
MCP is deliberately narrower than a general tool marketplace. It makes Atlas safely available to Apex while keeping identity, authorization, project scope, and persistence enforceable.
Meaningful memory
Persist verified facts, decisions, outcomes, and checkpoints rather than raw model output.
Behavioural guidance
Author and inspect project rules bound to semantic request regions rather than keywords.
Canonical scope
Derive private Atlas identities and memory boundaries from verified tenant, user, and project claims.
Controlled checkpoints
Consolidate related memories and commit state before handoffs or approved high-impact work.
Visible readiness
Separate service liveness from authenticated Atlas readiness so failures remain diagnosable.
Architecture
One clear responsibility boundary.
Ordinary session, record, and deterministic storage requests can remain within protected platform services. Every AI request crosses Apex; MCP is the governed Atlas command path inside that orchestration boundary.
Control planeEventium API
Authenticates actors, resolves entitlement and scope, classifies intent, and routes the request.
OrchestrationEventium Apex
Plans, approves, recalls permitted evidence, selects reasoning or execution, and streams results.
Memory gatewayEventium MCP
Checks caller authority and translates only allowlisted, identity-scoped Atlas operations.
Memory authorityAtlas
Stores and recalls facts, relationships, time, rules, provenance, and connected project context.
Reasoning + actionReady Gaia
Composes evidence-bounded results and uses an isolated workspace only when approved tools are required.
Browsers and execution workspaces never become memory authorities.They receive scoped results and evidence from the control plane; they do not receive Atlas credentials, private memory names, or unrestricted MCP access.
Allowlisted surface
Seven focused Atlas tools.
Every other tool is denied in the hosted production profile. MCP discovery publishes descriptions and schemas so approved clients know when—and when not—to invoke each operation.
Recall project memory
memory_recallRetrieve scoped, durable project evidence together with relevance signals and matching behavioural rules.
Atlas · recallStore durable memory
memory_storeRecord a verified fact, decision, result, or checkpoint with optional provenance and indexing aids.
Atlas · recordReview recent learning
memory_recentReturn a bounded view of recent project memories and report whether consolidation is due.
Atlas · recentConsolidate related memories
memory_consolidateCreate grouped summaries with source provenance while preserving the original project evidence.
Atlas · consolidateCreate a durability checkpoint
memory_persistCommit project memory and behavioural rules before a handoff or approved high-impact operation.
Atlas · checkpointAuthor a behavioural rule
rule_addBind a project rule from a natural-language example and directive inside the authenticated scope.
Atlas · rule recordList project rules
rules_listReturn the bounded set of behavioural rules attached to the authenticated project.
Atlas · rule recallSecurity
Identity proof is not permission.
Every hosted call fails closed. A valid credential identifies the caller; active entitlement, project membership, and an allowed role authorize the requested namespace and operation.
Interactive clients use a signed user credential. Its immutable subject is bound to the target identity and cannot select another user namespace.
Managed clients may use a hashed, revocable key linked to one user and checked against the same entitlement and project rules.
The private service credential authenticates Apex, while MCP still verifies the target user, role, project, and requested capability.
Scope is protected transport metadata.Tenant, user, project, and session identity are derived from authenticated context. A model cannot change them through tool arguments.
Client setup
Connect over Streamable HTTP.
Use a compatible MCP client and the standard initialize, discovery, and tool-call lifecycle. Send credentials in the Authorization header; never place secrets or identity scope in tool arguments.
Eventium MCP client configuration
{
"mcpServers": {
"eventium-atlas": {
"url": "https://mcp.eventium.ai/mcp",
"transport": "streamable-http",
"headers": {
"Authorization": "Bearer ${EVENTIUM_MCP_TOKEN}"
}
}
}
}
Endpoint
https://mcp.eventium.ai/mcpDiscoveryInitialize the session before calling
tools/list. Only the seven hosted tools should be returned.Canonical contextTenant, user, project, and session scope must originate in verified application context.
Evidence handlingRecall returns scoped evidence for Apex and Gaia. A memory record operation does not masquerade as an answer.
Use the documented lifecycle.The MCP integration guide explains initialization, session handling, discovery, and allowlisted invocation through the fixed hosted endpoint. The seven Apex-facing Atlas tools remain service-scoped; the browser never receives their credentials or an arbitrary MCP target.
Operations
Supported service references.
These endpoints expose only the operational metadata needed by clients and deployment checks. Authoring studios and unrestricted administrative surfaces are not part of the public gateway.