Skip to main content

D-04 MCP Gateway

Brick D: Tools, skills, plugins, MCP, connectors. French version: D-04-mcp-gateway

1. Role and responsibilities​

JSON-RPC 2.0 proxy to MCP servers, with the same controls as the Tool Gateway.

  • Relay initialize, tools/list, tools/call, resources/list, resources/read, prompts/list
  • Filter tools/list by allow list
  • Policy + DLP on tools/call

2. Position in the architecture and interactions​

  • Callers: Agents, MCP clients (IDE, assistants).
  • Execution path: Synchronous (JSON-RPC).
  • 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: Proxy, auth injection, controls.
  • 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
  • eai.tool_invocations

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​

  • POST /v1/mcp/gateway/{server_id}

Request POST /v1/mcp/gateway/{server_id}:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "search_issues",
    "arguments": {
      "q": "bug"
    }
  }
}

Response:

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "3 issues"
      }
    ]
  }
}

7. SDK examples​

Python:

from enterprise_ai import EnterpriseAI

client = EnterpriseAI(base_url="<control-plane-url>", api_key="eai_xxx_secret")
res = client.mcp.rpc("mcp_2", "tools/call", {"name": "search_issues", "arguments": {"q": "bug"}})

TypeScript:

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

const client = new EnterpriseAI({ baseUrl: "<control-plane-url>", apiKey: process.env.EAI_API_KEY });
const res = await client.mcp.callTool("mcp_2", "search_issues", { q: "bug" });

Java:

EnterpriseAIClient client = EnterpriseAIClient.builder()
    .baseUrl("<control-plane-url>")
    .apiKey(System.getenv("EAI_API_KEY"))
    .build();
JsonNode res = client.mcp().rpc("mcp_2", Map.of("jsonrpc", "2.0", "id", 1, "method", "tools/call", "params", Map.of("name", "search", "arguments", Map.of("q", "INC-42"))));

Go:

c, _ := eai.NewClient(eai.WithBaseURL("<control-plane-url>"), eai.WithAPIKey(os.Getenv("EAI_API_KEY")))
res, _ := c.MCP.Rpc(ctx, "mcp_2", "tools/call", params)

8. Configuration​

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

9. Security controls​

  • Scope tools:invoke
  • DLP on resources/read
  • 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_calls_total{server,method}, eai_mcp_latency_seconds
  • 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.call

11. FinOps units​

  • MCP calls counted as tool_call.
  • Emitted as Usage Event (POST /v1/finops/usage) with tenant, project, agent and run.

12. Failure modes and resilience​

  • JSON-RPC error propagated with its code
  • Timeout -> error -32000
  • Normalized errors (error envelope, message_i18n en/fr)

13. Standalone adoption (client of a client)​

  • enterprise_ai.mcp + Control Plane.
  • 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/call of an unallowed tool -> JSON-RPC error
  2. Cross-tenant access refused
  3. Error returned in fr with Accept-Language: fr

Kit L3 proof​

Governed applications: a credential bound to an application (third party agent) is refused when the application is not active or the MCP server is not granted to it (application_suspended, application_tool_not_granted), see the governed applications page.