Model Context Protocol in AACFlow: Connect Any Tool to Any AI Model

AACFlow has first-class MCP support built into the platform, not bolted on as an afterthought.

Supported transports:

• HTTP — standard REST-based MCP server communication
• SSE (Server-Sent Events) — for streaming tool responses in real time

This means your MCP server can run anywhere: on your laptop during development, inside your VPC, or on a public endpoint.

How to add a custom MCP server in AACFlow

Adding an MCP server to your workspace takes under two minutes:

1. Go to Settings in your AACFlow workspace.
2. Select MCP Servers from the left navigation.
3. Click Add MCP Server.
4. Enter a name, the server URL, and select the transport type (HTTP or SSE).
5. Save. The server is immediately available to all Agent blocks in your workspace.

Once added, Agent blocks automatically discover the tools exposed by that MCP server. You do not need to configure tools individually. The agent reads the server's tool schema at runtime and uses whichever tools are relevant to the current task.

For servers that require authentication, AACFlow passes credentials from your workspace credential store. You do not hardcode secrets in the server URL.

Workflow MCP servers: expose your workflows as tools

This is where AACFlow goes beyond standard MCP clients. Not only can AACFlow agents consume MCP tools — AACFlow can expose your workflows as MCP tools that other agents can invoke.

Every workflow you publish as an MCP server gets a stable endpoint. Other agents — whether they run in AACFlow or in an external system — can discover and call that workflow as a tool through the MCP protocol.

Practical example: you build a «Competitor Research» workflow in AACFlow that takes a company name and returns a structured analysis. You publish it as an MCP server. Now, any other agent in your organization — a sales assistant, a product planning agent, a strategic planning bot — can call that research workflow as a tool, without knowing how it works internally.

This creates a composable layer of organizational intelligence.

Real use cases for MCP in AACFlow

Database access. Connect a PostgreSQL MCP server to give your agents read (or write) access to your production database. The agent can query customer data, look up orders, or fetch inventory levels — all through a controlled, schema-validated interface.

Internal APIs. Your company probably has internal APIs that are not in AACFlow's catalog of 300+ integrations. An MCP server wrapping those APIs takes an afternoon to build and gives every agent permanent access.

Browser and web automation. MCP servers for Playwright or Puppeteer let agents browse the web, fill forms, and extract data from pages that don't have APIs.

File system access. Development agents that need to read files from a specific directory, check log files, or write outputs to a shared location can use a filesystem MCP server scoped to the relevant paths.

Vector databases. Pinecone, Qdrant, and Weaviate all have MCP servers. Connect them to give agents real-time retrieval-augmented generation without building a full RAG pipeline from scratch.

Step-by-step: adding a PostgreSQL MCP server

Here is a concrete example of connecting a PostgreSQL database to your AACFlow agents using an MCP server:

Step 1: Run the MCP server

Start the official PostgreSQL MCP server with your database connection string:

The server starts on a local port (default 3000) and exposes tools for running queries and inspecting the schema.

Step 2: Register in AACFlow

In your workspace Settings → MCP Servers, add:

• Name: Production PostgreSQL
• URL: http://localhost:3000 (or your deployed server URL)
• Transport: HTTP

Step 3: Use in an Agent block

Add an Agent block to your workflow. In the Tools section of the Agent block configuration, your PostgreSQL MCP server will appear automatically. Enable it, and the agent can now query your database to answer questions, generate reports, or trigger actions based on data.

The agent sees tools like query, list_tables, and describe_table. It chooses which tools to use based on the task — no additional configuration required.

Security considerations

MCP in AACFlow runs under the same permission model as the rest of the platform. Each workspace controls which MCP servers are available, and Agent blocks must explicitly enable a server's tools to use them.

For production deployments, we recommend running MCP servers inside your VPC rather than on public endpoints. AACFlow's networking proxies outbound requests, so your MCP server never needs to expose itself to the internet to be usable from AACFlow.

Why open standards win

AACFlow could have built a proprietary tool connectivity system. We chose not to, for the same reason we built A2A support alongside MCP: proprietary protocols create moats that hurt users.

With MCP, the tool you build for AACFlow works with Claude's API, with OpenAI's API, with any compliant client. Your investment in a well-designed MCP server compounds across every AI system in your organization.

AACFlow is the best runtime for MCP-powered workflows. But the tools you build are yours, and they travel wherever the protocol goes.

— Alexandr Chibilyaev