AI Access (MCP)
Let an AI agent such as Claude Code or Claude Desktop read your history, change settings and use Replay.
AI Access (MCP)
InterceptSuite has a built-in MCP server so an AI agent can work with the app: read Proxy History, change settings, use Replay and, if you allow it, work the Intercept queue.
It is off by default. Nothing listens until you turn it on.
Turn it on
Go to the top-level Settings tab > AI Access (MCP):
- Tick Enable AI access. The server listens on
http://127.0.0.1:7331/mcp(this computer only; you can change the port). - Copy the token, or click Copy the Claude Code command. An agent must present the token. Regenerate disconnects every agent.
- Choose what the agent may do.
These settings are per user and are not stored in project files, so a project someone sends you can never switch AI access on.
Connect an agent
Claude Code
claude mcp add --transport http interceptsuite http://127.0.0.1:7331/mcp --header "Authorization: Bearer <token>"
Claude Desktop (through the mcp-remote bridge):
{ "mcpServers": { "interceptsuite": {
"command": "npx",
"args": ["-y", "mcp-remote", "http://127.0.0.1:7331/mcp", "--header", "Authorization: Bearer <token>"] } } }
What the agent may do
| Permission | Default | Allows |
|---|---|---|
| Read | On | Search history (same filter syntax as the search box), view and decode packets, connections, logs and settings |
| Configure | On | Proxy settings, interception rules and switches, Scope, pass-through, clear history |
| Replay | On | Open sessions, send packets and read the replies |
| Replay only to Scope | On | Refuse Replay targets that are not in your Scope list |
| Intercept | Off | Forward, drop or modify held packets |
A permission that is off hides its tools from the agent. Changes apply immediately. An agent can use Replay sessions but never closes your tabs.
Safety
- This computer only. The server accepts connections from this machine alone, and every request needs the token.
- Everything is logged in the Logs tab.
- Captured traffic is untrusted. A packet can contain text aimed at the agent. Keep Intercept off unless you are watching, and keep Scope filled in.
- What the agent reads goes to its model provider. Captured traffic can contain passwords and tokens. Switch Read off, or leave AI access off, for traffic you cannot share.