Skip to main content

D-03 MCP Registry

Brick D: Tools, skills, plugins, MCP, connectors. French version: D-03-mcp-registry

1. Role and responsibilities​

References approved Model Context Protocol servers and their discovered tools.

  • Register a server (transport, url, auth)
  • Discover tools via tools/list
  • Per server tool allow list

2. Position in the architecture and interactions​

  • Callers: Portal, manifests (spec.mcp_servers).
  • Execution path: Synchronous.
  • Every interaction goes through the Control Plane API v1 (/v1) behind the API Gateway (D-14).

3. Open source component and developed part​

  • Open source: MCP SDK
  • Developed: Registry, discovery, approval.
  • External integration through an adapter eai/integrations/<system>.py with an in-process fallback when the URL is not configured.

4. Data model​

Tables (schema eai, tenant_id RLS):

  • eai.mcp_servers

Common columns: id uuid, tenant_id uuid NOT NULL, created_at, updated_at; RLS policy current_setting('app.tenant_id', true)::uuid.

5. Contract​

  • Applied contract: agent-manifest
  • JSON Schema: contracts/v1/agent-manifest.schema.json
  • Key fields: spec.mcp_servers[]
  • Validation: POST /v1/contracts/{name}/validate; manifest apply: POST /v1/manifests/apply.

6. API endpoints​

  • GET /v1/mcp/servers
  • POST /v1/mcp/servers
  • DELETE /v1/mcp/servers/{id}
  • GET /v1/mcp/servers/{id}/tools

Request POST /v1/mcp/servers:

{
  "name": "github",
  "transport": "streamable_http",
  "url": "https://<mcp-host>/github",
  "credential_ref": "secret/mcp/github",
  "allowed_tools": [
    "search_issues"
  ]
}

Response:

{
  "id": "mcp_2",
  "status": "active",
  "tools_discovered": 14
}

7. SDK examples​

Python:

from enterprise_ai import EnterpriseAI

client = EnterpriseAI(base_url="<control-plane-url>", api_key="eai_xxx_secret")
client.mcp.register_server(name="github", transport="streamable_http", url="https://<mcp-host>/github")

TypeScript:

import { EnterpriseAI } from "@enterprise-ai/sdk";

const client = new EnterpriseAI({ baseUrl: "<control-plane-url>", apiKey: process.env.EAI_API_KEY });
await client.mcp.registerServer({ name: "github", url: "https://<mcp-host>/github" });

Java:

EnterpriseAIClient client = EnterpriseAIClient.builder()
    .baseUrl("<control-plane-url>")
    .apiKey(System.getenv("EAI_API_KEY"))
    .build();
client.mcp().listServers().create("github", "https://<mcp-host>/github");

Go:

c, _ := eai.NewClient(eai.WithBaseURL("<control-plane-url>"), eai.WithAPIKey(os.Getenv("EAI_API_KEY")))
c.MCP.CreateServer(ctx, map[string]any{"name": "github", "url": "https://<mcp-host>/github"})

8. Configuration​

  • Environment variables: EAI_DATABASE_URL
  • Helm values keys: controlPlane.config, controlPlane.secrets

9. Security controls​

  • stdio servers forbidden outside the sandbox
  • Credential by OpenBao reference
  • Every request carries the tenant resolved from the credential; PostgreSQL RLS
  • Every mutation and policy decision writes a hash-chained audit event

10. Observability​

  • Prometheus metrics: eai_mcp_servers{status}
  • Traces: OpenTelemetry spans with GenAI semantic conventions (gen_ai.operation.name, gen_ai.agent.id, gen_ai.tool.name), attributes eai.tenant_id, eai.run_id, traceparent propagation.
  • Audit actions: mcp.server.register, mcp.server.delete

11. FinOps units​

  • No direct cost.
  • Emitted as Usage Event (POST /v1/finops/usage) with tenant, project, agent and run.

12. Failure modes and resilience​

  • Discovery failure: server in degraded state
  • Normalized errors (error envelope, message_i18n en/fr)

13. Standalone adoption (client of a client)​

  • enterprise_ai.mcp.
  • SDK: enterprise_ai.tools / mcp / skills / plugins / connectors / channels
  • Minimal deployment: Helm chart with controlPlane.enabled=true and PostgreSQL; disabled external components use their in-process fallback.

14. Acceptance tests​

  1. Tools outside the allow list not exposed
  2. Cross-tenant access refused
  3. Error returned in fr with Accept-Language: fr

Kit L3 proof​

Catalogue: POST /v1/mcp/servers/{server_id}/discover runs the MCP protocol (initialize, tools/list) and stores each tool with its input and output schemas, scopes, server and protocol versions (table eai.mcp_tool_versions, V156). The first version and every schema change open an approval (mcp_tool_version); GET /v1/mcp/catalog exposes versions and statuses (pending_approval, approved, rejected). Proof against a real MCP server (official Python SDK): tests/test_mcp_catalog_real.py.