Metantel · MCP Forge — agent tools for your business software

Your web apps,
safely in your agents' hands.

Give your AI agents the ability to act in the systems your team already runs — your ERP, CRM, helpdesk, or that homegrown admin panel. You choose exactly what to expose, describe it in plain English, and it goes live on your own application, with your own login and permissions. No rebuild. No new API to secure.

Your appyour auth, your permissions, your audit trail
Plain Englishdescribe a capability, review it, publish it
Stays healthymonitored and self-maintaining as your app evolves

What your agents can actually do

Not chat about your business — operate it. Every capability runs through the real application your people use, so what an agent does is exactly what a trained employee could do.

Use case

Automate back-office work

Let agents approve invoices, update records, and move tickets through your ERP, CRM, or helpdesk — through the same screens your team already uses.

Use case

Give staff copilots that act

An internal assistant that doesn’t just answer, but does the work: looks up an order, files a return, schedules the job — in your real systems, not a demo.

Use case

Expose safe actions to customers

Offer a support or sales agent a narrow, approved set of actions — and nothing beyond what you allow. You choose the boundary.

Use case

Use the software you already have

Legacy ERPs, homegrown admin panels, modern web apps — if your people run it in a browser, your agents can use it too. No rebuild.

Inside MCP Forge · the pipeline, in miniature

From a sentence to a server you own.

This is what actually happens when you build with Flow Console's MCP Forge. It isn't magic — it's a streamlined, low-overhead pipeline with an agent assisting at every step. Pick a capability or type your own, then run the forge. Nothing here touches a real system.

I want an agent that can…
  1. 01 Understand

    The agent reads your request and consults what Flow Console already knows about your application — its knowledge graph and precompiled documentation.

  2. 02 Map the workflow

    The workflow is pinned at the protocol level — the same typed requests your app already serves. No screen automation, nothing brittle.

  3. 03 Forge & verify

    Forge compiles a TypeScript MCP server with clean tool definitions, then verifies it end-to-end before you publish.

src/tools/your_tool.ts · generated by MCP Forge
Your server. Your call.
Self-host

Deploy the generated project inside your own perimeter. It authenticates as your application does, so it inherits your login, your permissions, and your audit trail.

Metantel-hosted · free

Or let us run it on Metantel's servers at no cost — isolated per tenant, credentials stay yours, revocable with one click.

Either way, agents only ever get the tools you approved — nothing more.

Secure by inheritance

Built on software your security team already trusts

The safest new interface is the one you don't build. Because agents act through your existing, battle-tested application, they're bound by every control that already protects it — automatically.

It reuses the security you already built

Agents act through your existing application, so they inherit its login, its permissions, its input validation, and its audit trail. There is no new, unvetted interface to harden.

No new attack surface

You aren’t opening a fresh database endpoint for AI. Every action takes the same path a signed-in employee’s click takes — through guardrails your team has hardened for years.

Acts with the caller’s own permissions

An agent can only do what the person or service behind it is already allowed to do. Least privilege comes for free from your application’s own roles.

Runs inside your perimeter, under your controls

It’s deployed and operated by you, on your infrastructure, using your identity provider and monitoring. Your security stack governs it like any internal service.

Lower engineering cost

No integration project. No dedicated team.

Connecting AI to real systems usually means a custom API layer, a roadmap, and people to maintain it. Here, your application is the interface and the tools are described, not coded.

No integration team required

You don’t build and maintain a bespoke API layer just so agents can act. Your existing application is the interface.

Describe tools in plain English

Say what you want an agent to be able to do; MCP Forge turns that sentence into a working, reviewable tool. You approve it, then it’s live.

Add or change capabilities anytime

A new workflow next quarter? Describe it and publish. No release train, no SDK version bumps, no re-integration.

Maintenance is largely automated

When your app changes, tools keep working — or you’re told exactly what needs a look. No standing team babysitting a brittle integration.

Health checks & maintenance

It keeps itself working as your app changes

Software moves. Fields get renamed, pages get redesigned. Instead of an agent breaking in front of a customer, the platform notices, adapts what it safely can, and tells you the rest.

Continuous health checks

Your published tools are watched around the clock to confirm they still do exactly what they should — safely, without side effects.

Change detected automatically

When your app is updated and something shifts under a tool, it’s caught immediately — not by an agent failing in front of a customer.

Self-heals where it safely can

Many app changes are absorbed automatically. Anything risky is paused and flagged for a quick human OK — nothing changes behavior silently.

You get the numbers

Coverage, uptime, and what needed attention — reported per application, so you always know the state of your agent surface.

You build it. You own it.

Adopted by your team, run inside your world

This isn't an outside system reaching into your software. You stand it up on your own infrastructure, decide what it can touch, and operate it with your own identity and monitoring — so it earns trust the same way any internal tool does, and your security team stays in control.

01

You choose what to expose

You decide exactly which capabilities become agent tools. Nothing is exposed that you didn’t explicitly approve.

02

Describe it or point to it

Pick an existing workflow, or describe a new one in plain language. MCP Forge drafts the tool for you to review.

03

Review and publish

You approve each tool and get a clean TypeScript MCP server. Run it on your own infrastructure — or let Metantel host it for free, isolated and revocable at any time.

04

Stay in control

Scope every agent to just the tools it needs, watch every action in your own audit trail, and switch anything off instantly.

Give your agents safe hands in your own software.

Turn the systems you already trust into a governed set of actions your agents can use — built in plain English, secured by your own application, and kept healthy for you.