Skip to main content

D-02 Tool Gateway

Brick D: Tools, skills, plugins, MCP, connectors. French version: D-02-tool-gateway

1. Role and responsibilities​

Single tool invocation point for agents: scopes, policy, DLP, approval, credential lease, cost, audit.

  • Validate input against input_schema
  • Evaluate OPA tool.invoke and its obligations
  • DLP on input and output
  • Obtain a lease (D-11) then call the endpoint or the sandbox

2. Position in the architecture and interactions​

  • Callers: Loop Controller (C-03), SDK.
  • Execution path: Synchronous; can switch to approval wait (durable).
  • 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: none (to develop)
  • Developed: Fully developed.
  • 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.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: tool-manifest
  • JSON Schema: contracts/v1/tool-manifest.schema.json
  • Key fields: spec.requires_approval, spec.rate_limit, spec.sandbox_profile, spec.credential_ref
  • Validation: POST /v1/contracts/{name}/validate; manifest apply: POST /v1/manifests/apply.

6. API endpoints​

  • POST /v1/tools/{name}/invoke
  • GET /v1/tools/invocations

Request POST /v1/tools/{name}/invoke:

{
  "input": {
    "q": "INC-42"
  },
  "run_id": "run_01J9",
  "step_index": 4
}

Response:

{
  "invocation_id": "tinv_77",
  "status": "succeeded",
  "output": {
    "issues": [
      {
        "key": "INC-42"
      }
    ]
  },
  "decision_id": "dec_51",
  "cost": {
    "amount": "0.001",
    "currency": "EUR"
  }
}

7. SDK examples​

Python:

from enterprise_ai import EnterpriseAI

client = EnterpriseAI(base_url="<control-plane-url>", api_key="eai_xxx_secret")
out = client.tools.invoke("jira.search", {"q": "INC-42"}, run_id="run_01J9")

TypeScript:

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

const client = new EnterpriseAI({ baseUrl: "<control-plane-url>", apiKey: process.env.EAI_API_KEY });
const out = await client.tools.invoke("jira.search", { arguments: { q: "INC-42" }, run_id: "run_01J9" });

Java:

EnterpriseAIClient client = EnterpriseAIClient.builder()
    .baseUrl("<control-plane-url>")
    .apiKey(System.getenv("EAI_API_KEY"))
    .build();
JsonNode out = client.tools().invoke("jira.search", Map.of("arguments", Map.of("q", "INC-42")));

Go:

c, _ := eai.NewClient(eai.WithBaseURL("<control-plane-url>"), eai.WithAPIKey(os.Getenv("EAI_API_KEY")))
out, _ := c.Tools.Invoke(ctx, "jira.search", map[string]any{"arguments": map[string]any{"q": "INC-42"}})

8. Configuration​

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

9. Security controls​

  • Scope tools:invoke + scopes_required
  • Secrets never exposed to the agent: lease injected gateway side
  • Approval if requires_approval or OPA obligation
  • 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_tool_invocations_total{tool,status}, eai_tool_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: tool.invoke, dlp.redact

11. FinOps units​

  • Tool calls (unit tool_call) and declared spec.cost.
  • Emitted as Usage Event (POST /v1/finops/usage) with tenant, project, agent and run.

12. Failure modes and resilience​

  • Non idempotent tool: no automatic retry
  • Timeout and circuit breaker per tool
  • Rate limit -> 429
  • Normalized errors (error envelope, message_i18n en/fr)

13. Standalone adoption (client of a client)​

  • enterprise_ai.tools + Control Plane; OPA and OpenBao optional (fallbacks).
  • 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. Invocation without scope -> 403 policy_denied
  2. Output containing an IBAN masked
  3. requires_approval -> state waiting_approval
  4. Cross-tenant access refused
  5. Error returned in fr with Accept-Language: fr