Skip to main content
Glama
mcp-server-make

mcp-server-make

by phlppgdfry

Make Enterprise Workbench

Validate workflows Production health

A traceable automation showcase by Philippe Godfroy: ask an actual Make knowledge agent, research approved company websites and inspect persistent execution records. The operational application and the synthetic reference laboratory are clearly separated.

Open the operational application · Health · Acceptance evidence · Reference lab

Actual capabilities

Capability

Operational behavior

Boundary

Repository knowledge

Browser → authenticated backend → existing Make Repo Kennisagent → validated citations

Seven public documents pinned to 885c72b6a247ef9f9aabe7bc540339888397f403; answers concern that historical snapshot

Company research

Actual HTTPS homepage metadata, robots checks, source excerpts and observation time

Approved domains only; no invented contacts, emails, fit scores or independently verified firmographics

Execution ledger

D1 persistence, request IDs, audit events, idempotency and uncertain-state handling

Single operator; seven-day retrieval period; physical retention purge is operator-triggered

Credit control

Server-side allocation, provider usage receipts and atomic reconciliation

Existing approved 50-credit allowance; no purchases or automatic model retries

Integration inspection

Read-only inspection of the real agent and existing scenarios

Detailed inventory and execution history require operator access

CRM

Explicit crm_unconfigured error

No configured CRM, no contact writes and no outbound communications

Seven fresh browser-to-Make acceptance cases passed, with 9.99 actual Make credits reported. Two further fresh release smoke tests used 1.76 and 1.88 credits: 13.63 credits total, below the approved 50. The latest local suite has 111 passing Node tests, plus the legacy local pattern experiments. These numbers do not establish a security certification or an enterprise SLA. The acceptance report records exact commits, deployment evidence and remaining prerequisites.

Related MCP server: Make MCP Server

Architecture

flowchart LR
  Browser[Vanilla HTML / CSS / JavaScript] --> API[Same-origin Cloudflare Worker]
  API --> Auth[Operator authentication / validation / rate limits]
  Auth --> Ledger[(D1 executions / audit / credit ledger)]
  Auth --> Make[Existing Make knowledge agent]
  Auth --> Research[Approved public company websites]
  Make --> Sources[Pinned quotation validation]
  Sources --> Ledger

The Worker serves both frontend and API through Sites. Secrets remain server-side. Public visitors can inspect health and capability status; model calls, research, history and maintenance require a separate operator key. The browser holds that key only in tab memory. Model output is rendered as text. A verified quotation proves provenance, not the truth of every claim in an answer.

The original GitHub Pages lab remains available with synthetic support, tenant-scoped knowledge retrieval and deterministic lead scoring. It does not use a model, live research or connected business accounts. Its Node gateway (scripts/server.js) is a separate local reference, not the operational backend.

Local setup

Node.js 22.13+, preferably Node 24, npm and a full Git clone are required. The build reads knowledge from the pinned historical commit.

git clone https://github.com/phlppgdfry/make-enterprise-showcase.git
cd make-enterprise-showcase
npm ci
npm run test:all
npm run build:production
npm run start:production

Open http://127.0.0.1:4179. The local preview has an in-memory SQLite database, the development key local-preview-only, and no Make credentials. Its data disappears on restart. It can fetch real approved public company websites when network access is available.

For the original synthetic lab, run npm start and open http://127.0.0.1:4173. SHOWCASE_API_TOKEN enables that reference gateway's authenticated fixture endpoints. It is distinct from the hosted operator key.

Production configuration

Setting

Purpose

MAKE_API_KEY

Existing authorized Make token; secret runtime storage only

MAKE_ZONE

Make API region; this deployment uses eu1.make.com

OPERATOR_API_KEY

Independent high-entropy operator bearer secret

RESEARCH_DOMAINS

Optional approved domains; defaults to make.com,cloudflare.com,github.com

DB

D1 binding provisioned from production/.openai/hosting.json

The owner explicitly approved secure runtime credential storage and at most 50 existing Make credits. No credentials are in source control. The locally generated operator key is in the ignored owner-readable exports/production-operator-key.txt; do not publish it. No CRM credential is configured.

Verification and deployment

npm run test:all          # 111 Node tests, local experiments and static blueprint audit
npm run format:check     # Production source formatting
npm run build:production # Bundle Worker and public assets
npm run smoke:production # Actual public checks; no metered AI calls
npm audit                # Dependency advisories

CI installs locked dependencies, tests, builds, checks formatting and audits dependencies. The public production health workflow runs every six hours and on manual dispatch. Sites publication is performed with authenticated source/archive tooling; a GitHub push alone does not redeploy the Worker. The artifact records its canonical GitHub source commit, which is returned by /api/health.

See deployment, environment setup and rollback, architecture and security boundaries, integration configuration, repository audit and release acceptance.

Existing Make resources

The existing agent is reused; no duplicate agent or scenario was created. Its seven knowledge documents and read-only permissions are preserved. Five existing scenarios were inspected without changing or executing them. The active webhook/datastore example is not a verified DLQ or exactly-once system; inactive support, CRM, ETL and knowledge scenarios remain templates requiring their own configuration.

The older blueprint deployment tooling still defaults to a plan and requires explicit selection and configuration for writes. The incomplete legacy datastore provisioner no longer claims successful schema provisioning. SDK sketches and earlier audit scores remain historical material, not supported production services.

Limits before real customer operations

CRM integration requires an authorized destination, credentials, sandbox fields and explicit approval for test writes. Research provides self-published website metadata, not a contact discovery service. The agent's snapshot intentionally predates the operational application: use current status/docs for current deployment facts. Physical retention maintenance remains manual at the owner’s request. Continuous customer operations also need an approved backup/restore process, access governance and operational ownership. Managed D1 point-in-time recovery and a cloud restore drill have not been verified. The showcase is publicly deployed and its working flows are verified; it is not represented as an unrestricted enterprise production system.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      D
      quality
      D
      maintenance
      Transform your Make scenarios into callable tools for AI assistants. Leverage your existing automation workflows while enabling AI systems to trigger and interact with them seamlessly.
      6
      110 npm
      174
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      This server enables AI assistants to discover, parameterize, and trigger Make.com automation workflows configured for on-demand execution. It allows users to execute scenarios and receive structured JSON results, bridging the gap between AI assistants and complex automation ecosystems.
      MIT