Open standards, end to end
Agents are only as governable as their connections. Every tool is an MCP server, identity is OIDC, telemetry is OpenTelemetry — and the model is a configuration field, not a commitment.
Model Context Protocol
Tools with a contract
MCP turns 'what can this agent touch?' from an investigation into a catalog query.
Every tool is an MCP server
ShippedAgents reach data and actions exclusively through Model Context Protocol servers — a typed JSON-RPC contract, not ad-hoc plugins. What an agent can touch is enumerable, reviewable and testable.
Three governed namespaces
ShippedProduct tools (aro.*, dh.*), shared platform tools, and tenant-specific tools — each namespace has its own ownership, review path and deployment cadence.
Production-grade tool kit
ShippedPer-run JWT auth, idempotency keys, SLA timeouts and OpenTelemetry spans are part of the in-repo MCP server kit — tools are infrastructure, not scripts.
Bring your own MCP servers
GA Q3 2026The same contract that connects NeoXam products connects yours: in-house systems and third-party platforms join the catalog as tenant-scoped MCP servers, governed by the same policies.
NeoXam suite
Embedded across the products that run your books
Statuses are exact — shipped connectors are contract-tested today; live product connections arrive with Beta and GA.
Aro
Shipped — pilotReconciliation positions, match history and break context power the reconciliation investigator — the platform's shipped pilot.
DataHub
Shipped — task modeBusiness-rule corpus, data-quality context and rule execution back the five-agent Business Rules family.
GP
GA Q3 2026Investment accounting and NAV context arrive with the GP connectors at general availability.
Impress
GA Q3 2026Reporting datasets and document context connect for reporting assistance at general availability.
Density
GA Q3 2026Portfolio and compliance context from the PMS connects for monitoring use cases at general availability.
Standards
Nothing proprietary where it matters
Identity, telemetry, schemas and model routing all follow open standards — adopting the platform never means abandoning your stack.
OIDC for identity
BetaKeyCloak, Azure Entra ID and Okta-compatible single sign-on with Authorization Code + PKCE — your IdP remains the source of truth.
OpenTelemetry for traces
ShippedEvery run emits standard OTel spans; ship them to Langfuse or your own observability stack. No proprietary trace format.
Model-agnostic by design
Anthropic todayThe model is a configuration field in the agent descriptor, routed through a gateway. Claude (Anthropic) is the supported model today; additional providers are qualified through the same eval gates as they are certified.
Schema-validated I/O
ShippedAgent inputs and outputs are JSON-Schema typed. Downstream systems consume structured results, not free text.
FAQ
Frequently asked questions
Why MCP rather than a proprietary plugin framework?
Because tool access is the security boundary of an agent platform. MCP gives that boundary an open, typed, vendor-neutral contract: every capability an agent can reach is a declared server with enumerable tools, which makes review, testing and audit tractable — and keeps your integration work portable.
Can agents call systems outside the NeoXam suite?
That is the design: tools are MCP servers, so any system wrapped in an MCP server can join the catalog under tenant scoping and policy control. Today the shipped connectors cover Aro and DataHub scenarios; tenant-provided MCP servers are a GA capability.
Which LLM providers are supported?
The descriptor treats the model as configuration, routed through a gateway. In production today the platform runs Claude (Anthropic). Additional providers are qualified through the same evaluation baselines before they are offered — model choice never bypasses the eval gate.
Do the shipped connectors run against live product data?
V0 connectors are contract-tested against fixture-backed implementations of the product APIs; live Aro and DataHub connections arrive with the Beta. The MCP contract is identical in both cases, which is the point of testing against contracts.
How do integrations affect what an agent may do?
Tool access is governed per agent by allow / ask / deny policies and per environment by network policy. An agent only sees the MCP servers its descriptor declares and its environment permits — integration never widens permissions implicitly.
Keep exploring
More of the platform
- Platform overviewThe governed agentic platform, end to end.
- Agents & catalogDeclared, versioned agents — descriptors, skills, modes.
- Governance & evalsPR review, eval baselines, allow/ask/deny policies.
- Security & trustIdentity, tenant isolation, vaults, append-only audit.
- DeploymentEU SaaS today; hybrid, on-prem and sovereign at GA.
- Observability & costOpenTelemetry traces, budgets, cost per run and tenant.
- ArchitectureRunner, harness, control plane vs execution plane.
General availability comes in Q3 2026. The Early Adopter Program is open now.
A limited cohort, a one-year platform trial and three workshop streams — Business ROI, Compliance, Operational fit. Bring one workflow; leave with a governed agent and the evidence to certify it.