Built-in MCP Support

RelayCraft features Built-in MCP (Model Context Protocol) Support, providing a bridge between your live network traffic and AI agents. AI assistants like Claude Desktop, Cursor, or any MCP-compatible client can directly “see” your requests, allowing for real-time traffic analysis, automated rule creation, and interactive debugging.

Enabling the MCP Server

1. Open Settings → Integrations → MCP Server.
2. Toggle it on. RelayCraft starts an HTTP server on the configured port (default 7090).
3. Copy the Bearer token shown in the panel — you’ll need it for write operations.

Connecting a Client

Configure your MCP client to point to:

http://localhost:7090/mcp

For Claude Desktop, add an entry to claude_desktop_config.json:

{
  "mcpServers": {
    "relaycraft": {
      "type": "http",
      "url": "http://localhost:7090/mcp",
      "headers": {
        "Authorization": "Bearer <your-token>"
      }
    }
  }
}

The server uses JSON-RPC 2.0 over HTTP POST and follows the MCP protocol spec (2024-11-05).

Available Tools

Read Tools

These tools do not require authentication and are safe to expose to any MCP client.

list_sessions

List all recorded debugging sessions.

list_sessions()
→ Session IDs, names, timestamps, flow counts

list_flows

Query flows in a session with optional filters.

Parameter	Type	Description
session_id	string?	Session to query. Omit for the active session.
limit	integer	Max flows to return. Default 50, max 200.
method	string?	Filter by HTTP method: GET, POST, etc.
status	string?	Filter by status code or range: 404, 4xx, 5xx.
domain	string?	Substring match on the host.
has_error	boolean?	Return only flows that have errors.
content_type	string?	Substring match on the response Content-Type.

get_flow

Get full details of a single flow — headers, request body, response body.

Parameter	Type	Description
id	string	Flow ID from list_flows or search_flows.

Response bodies larger than 100 KB are truncated.

search_flows

Search flows by keyword. By default, matches against the full URL. Use search_in to search inside request/response bodies or headers instead.

Parameter	Type	Description
query	string	Keyword to search for.
search_in	string	Where to search: url (default), response_body, request_body, header.
case_sensitive	boolean	Case-sensitive match. Default false.
session_id	string?	Session to search within.
limit	integer	Max results. Default 20, max 50.

Body and header searches scan up to the 5 000 most recent flows. URL search is instant.

# Find all flows whose response body mentions "NullPointerException"
search_flows(query="NullPointerException", search_in="response_body")

# Find flows passing a specific token in a header
search_flows(query="Bearer eyJ", search_in="header")

get_session_stats

Aggregate statistics for a session — total flows, error rate, top domains, status distribution, slowest requests.

Parameter	Type	Description
session_id	string?	Session to summarise. Omit for the active session.

list_rules

List all rules with key fields (ID, name, type, URL pattern, enabled state, source).

list_rules()
→ Rule list for follow-up actions like delete/toggle

---

Write Tools

Write tools require the Authorization: Bearer <token> header.

replay_request

Replay a captured request through the proxy, optionally with modifications. The replayed request appears in the traffic list so you can inspect the response with get_flow.

Parameter	Type	Description
flow_id	string	Flow to replay.
modifications.url	string?	Override the full URL including scheme, host, path, and query string.
modifications.method	string?	Override the HTTP method.
modifications.headers	object?	Key-value pairs to add or override request headers.
modifications.body	string?	Replacement request body.

# Replay with a different endpoint
replay_request(
  flow_id="abc123",
  modifications={
    "url": "https://api.example.com/v2/users",
    "headers": {"Authorization": "Bearer new-token"}
  }
)

create_rule

Create a proxy rule that takes effect immediately.

Parameter	Type	Description
type	string	Rule type (see below).
name	string	Descriptive rule name.
url_pattern	string	URL substring to match.
method	string?	Optional HTTP method filter.

Rule types and their extra parameters:

Type	Extra Parameters
map_local	mock_body, mock_content_type (default application/json), mock_status (default 200)
map_remote	target_url (required)
rewrite_body	rewrite_target (request/response), rewrite_mode (set/replace/regex_replace/status_code), rewrite_content, rewrite_pattern, rewrite_replacement, rewrite_status, rewrite_content_type
rewrite_header	header_phase (request/response), header_operation (add/set/remove), header_name, header_value
throttle	bandwidth_kbps, delay_ms
block_request	(no extra params)

delete_rule

Delete a rule by ID.

Parameter	Type	Description
rule_id	string	ID of the rule to delete.

toggle_rule

Enable or disable a rule without deleting it.

Parameter	Type	Description
rule_id	string	Rule to toggle.
enabled	boolean	true to enable, false to disable.

Example Workflow

Here is a typical debugging session driven by an AI agent:

1. Capture traffic — run your app while RelayCraft is intercepting.
2. Ask the AI: “Find all 5xx errors from api.example.com” → list_flows(domain="api.example.com", status="5xx")
3. Ask: “Show me the full request and response for the first error” → get_flow(id="...")
4. Ask: “Replay it but change the Authorization header” → replay_request(flow_id="...", modifications={headers: {...}})
5. Ask: “Mock that endpoint to return a fixed 200 response while I fix the backend” → create_rule(type="map_local", ...)