MCP Registry
Eight MCP servers expose Nexora AI's back-office systems as tools - for agents, assistants, and anything else that speaks the Model Context Protocol.
Why a Registry
A language model doesn't need an interface description, it needs tools: clearly named calls with clearly named parameters. The Model Context Protocol (MCP) is the format for that.
The servers in this registry don't exist alongside the APIs, they're built from them. If you know the APIs, you'll find the same operations here - just as a tool instead of an endpoint. What applies to the API catalog therefore applies here too: the same error format, the same authentication, the same permissions.
The Kong AI Gateway sits in front either way. It decides who may call which server and which tool, logs every call, and keeps the tool list current.
MCP Servers for Developers
This list comes directly from the Konnect MCP registry nexora-ai-ops. It always shows the published state - when a server is added, it appears here without this page needing to change.
Access & Onboarding Requests
1.0.0
Manage access requests, find capability profiles that match an agent.
Agent Deployments
1.0.0
List, retrieve, register and update AI agents in the deployment inventory.
Usage Chargeback & Settlement
1.0.0
Retrieve usage settlements and pre-calculate the chargeback.
Compliance Registry
1.0.0
Query external regulatory and compliance policy definitions.
Cost & ROI Estimation
1.0.0
Determine cost estimates, comparable deployments and usage-index figures for an agent.
Evaluation Sessions
1.0.0
Schedule and reschedule evaluation sessions, check open time slots and log feedback.
Model Registry Extract
1.0.0
Retrieve provenance and lineage records for a foundation model.
Pricing Benchmark Index
1.0.0
Query external LLM/agent unit-cost benchmarks by provider and model tier.
Nexora AI Back-Office Systems
These servers expose Nexora AI's own APIs as tools. Each entry is reachable at its path on the gateway.
| Server | Path | Tools |
|---|---|---|
| Agent Deployments | /mcp/agent-deployments | list-agentsget-agentregister-agentupdate-agentlist-agent-artifacts |
| Access & Onboarding Requests | /mcp/access-requests | list-requestsget-requestlog-requestget-request-scorematching-agents |
| Evaluation Sessions | /mcp/evaluation-sessions | list-sessionscreate-sessionreschedule-sessionavailable-timeslotslog-feedback |
| Cost & ROI Estimation | /mcp/cost-estimation | request-estimateget-estimatequery-comparablesquery-usage-index |
| Usage Chargeback & Settlement | (tool library) | list-settlementsget-settlementcalculate-chargeback |
Usage Chargeback & Settlement is deliberately set up without its own endpoint. It serves as a library that other servers draw their tools from - settlement data shouldn't accidentally end up on an open endpoint.
Connected Third-Party Systems
These servers already speak MCP. The gateway passes them through unchanged and adds access control and logging.
| Server | Path | What it's for |
|---|---|---|
| Pricing Benchmark Index | /pricing-benchmark-index | External index of LLM/agent unit-cost benchmarks by provider and model tier |
| Model Registry Extract | /mcp/model-registry-extract | Provenance and lineage records for third-party foundation models |
| Compliance Registry | /mcp/compliance-registry | External regulatory and compliance policy lookups |
Third-party systems get the same access control as Nexora AI's own servers. A key that isn't allowed to see model-provenance data doesn't even get that tool shown in its list.
From the Spec, Not Beside It
The back-office tools are generated from the same OpenAPI specifications that feed the API catalog. When an endpoint changes, the tool changes with it - there's no second description that can drift out of date.
Permissions per Tool
Access is controlled not just per server, but per tool. A read-only key gets list-agents but not update-agent - and sees the difference right in the tool list.
Every Call Is Traceable
The gateway logs which key called which tool when, and with what result. For an agent that decides on its own, that's not a nice-to-have, it's the prerequisite.
Access: MCP servers are called with the same API key as the other interfaces. To learn how to generate a key, see Authentication. The gateway's base address depends on the environment - the paths above stay the same across all environments.
Server Operating Modes
An MCP server in the gateway is always one of several operating modes. Three of them are in use for Nexora AI:
| Mode | What it does | In use here |
|---|---|---|
conversion-listener | Translates a REST API into tools and accepts MCP requests | The four back-office systems |
conversion-only | Translates a REST API into tools without being reachable itself | Usage Chargeback & Settlement |
passthrough-listener | Passes a third-party MCP server through without changing its tools | The three third-party systems |
If you want to rebuild the servers yourself, the full declarative configuration is in the repository under ai-gateway/config/mcp-servers.yaml.
Ready for Your First Integration?
Register, create an application, and generate your first catalog entry in under ten minutes.