MCP AI setup

Connect an MCP-compatible client in minutes.

Use the local Streamable HTTP endpoint for the simplest setup, or the per-installation HTTPS endpoint when the client requires TLS. Both expose the same installed MCP tool catalog.

Connection

Use the local Streamable HTTP endpoint

Social Browser exposes the same MCP AI capability catalog over local HTTP and HTTPS.

HTTP — recommended

Maximum compatibility for a client on the same computer.

http://127.0.0.1:60080/mcp

HTTPS — TLS 1.2+

Encrypted local transport. Strict clients may need the installation public certificate.

https://127.0.0.1:60081/mcp

Configuration

Generic client configuration

Client products use different config file names, but the MCP server concept is the same.

{ "name": "social-browser", "transport": "streamable-http", "url": "http://127.0.0.1:60080/mcp" }

After an update: reconnect or restart the MCP client so it reloads the current tool catalog. Many clients cache tool definitions for the connection lifetime.

Setup

Connection checklist

Most connection problems come from transport, certificate or stale-catalog issues rather than the browser workflow itself.

1. Enable MCP AI

Open Social Browser MCP AI settings and confirm the intended operation policies.

2. Choose transport

Start with local HTTP. Use HTTPS only when the client requires a secure URL.

3. Discover tools

Use tools/list / Agent Discovery and do not hard-code stale tool assumptions.

4. Respect policy

Allow, Ask or Block is controlled by the device owner at operation level.

5. Resolve targets

Use stable Profile identity and rediscover runtime tabs/windows after reconnect.

6. Diagnose first

If a long task fails, inspect capability truth and readiness before retrying blindly.

Security

Security model

The endpoints are local-only by default. Remote access requires explicit server binding and network security configuration; changing the IP in a client does not enable remote access.

Do not expose the local MCP endpoint directly to an untrusted network. Use the browser permission model and explicit network controls when building remote workflows.