Keep & automate
MCP: let AI agents use Akev
Free and Pro. The read tools are free and unlimited (except
akev_llm_conversation, part of the Pro LLM traffic lens). Tools that change something (send or replay requests, set rules, add mocks, resume breakpoints, start and stop recording, clear traffic, activate interceptors) need Akev Pro or its 14-day trial. Without it they return an error naming the feature. The table shows which is which.
Akev includes a Model Context Protocol server, so coding agents can read captured traffic, search it, replay requests and create mocks.
Connect
MCP clients start akev mcp and talk to it over stdio. akev mcp connects to the Akev that
is running, the desktop app or akev start, and reconnects if Akev restarts during the
session. When no Akev is running, it starts a headless one, which stops when the client ends
the session.
akev mcp is part of the akev command-line tool (download). The desktop
app doesn't install the akev command, so to connect agents to the desktop app, install the
command-line tool as well.
If you start Akev with --data-dir, give akev mcp the same one
("args": ["mcp", "--data-dir", "/path/to/dir"]) so it finds that instance.
HTTP. A client that connects over HTTP (Streamable HTTP) can send its requests to
POST /mcp on Akev's API address, with Authorization: Bearer <token>. For akev start that
is http://127.0.0.1:45456/mcp, or another port when 45456 is busy (Settings → About →
API shows the address). The token is random on every run unless you set one with --token
or AKEV_TOKEN. The desktop app picks a random port and token each time it starts, so connect
to it through akev mcp.
Set up your client
The quickest way: in Akev, open Settings → AI agents (MCP). It has a ready-to-copy
setup for Claude Code (the claude mcp add command, optionally with --scope user),
Claude Desktop (its JSON, with the config file's path on each OS), Cursor and
VS Code (their mcp.json, and install links), either through akev mcp (recommended; the
path to akev is editable for apps that don't see your shell's PATH) or over HTTP with the
/mcp address and token (masked on screen, real in what you copy). It also lists the tools,
which are free and which need Pro, and the safety switches below.
The same by hand: every client runs the same command, akev with the argument mcp. Apps
started from a desktop menu may not see the PATH of your shell. If a client can't start
akev, give the full path instead: which akev (macOS, Linux) or where akev (Windows)
prints it.
Claude Code
claude mcp add akev -- akev mcpThis adds Akev to the current project, for you only (scope local, the default). Add
--scope user for all your projects, or --scope project to share it with your team through
.mcp.json in the project:
claude mcp add --scope user akev -- akev mcpclaude mcp list shows whether Claude Code can connect.
Claude Desktop
Open Settings → Developer → Edit Config, or edit claude_desktop_config.json directly:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"akev": { "command": "/usr/local/bin/akev", "args": ["mcp"] }
}
}Use the full path to akev: Claude Desktop may not see your shell's PATH. On Windows, double
the backslashes, as in "C:\\Tools\\akev.exe". Quit and reopen Claude Desktop to load the
change.
Cursor
Add Akev to ~/.cursor/mcp.json for all your projects, or to .cursor/mcp.json in a project:
{
"mcpServers": {
"akev": { "command": "akev", "args": ["mcp"] }
}
}VS Code
Add Akev to .vscode/mcp.json in your workspace. VS Code uses servers, not mcpServers:
{
"servers": {
"akev": { "type": "stdio", "command": "akev", "args": ["mcp"] }
}
}Tools
| Tool | Purpose | Plan |
|---|---|---|
akev_status |
Proxy address, CA, stats, and the environment variables that route a program through Akev | Free |
akev_list_exchanges |
List exchanges, optionally with a filter | Free |
akev_get_exchange |
Full request/response with decoded bodies | Free |
akev_get_body |
Read a request or response body (or a slice of it) as text, formatted or base64 | Free |
akev_search |
Full-text search of URLs, headers and bodies | Free |
akev_wait_for |
Wait until an exchange matching a filter is captured | Free |
akev_get_rules |
Read the rules, with match counts | Free |
akev_test_rules |
Dry-run rules against a captured exchange or an ad-hoc request | Free |
akev_export_har |
Export traffic as HAR, inline or to a file | Free |
akev_snippet |
Generate code for a captured request | Free |
akev_list_breakpoints |
Exchanges paused at a breakpoint | Free |
akev_list_interceptors / akev_deactivate_interceptor |
List clients Akev can launch or configure; undo one | Free |
akev_filter_help |
The filter language reference | Free |
akev_llm_conversation |
The conversation of an LLM API call: messages, tool calls, reply, tokens, latency | Pro (LLM lens) |
akev_send_request |
Send a request (like Compose) | Pro |
akev_replay |
Replay a captured request, optionally modified | Pro |
akev_set_rules |
Replace the rules | Pro |
akev_add_mock |
Quickly add a mock rule | Pro |
akev_resume_breakpoint |
Resume a paused exchange, optionally edited | Pro |
akev_set_recording |
Pause or resume recording | Pro |
akev_clear_traffic |
Clear captured traffic | Pro |
akev_activate_interceptor |
Launch or configure a client (browsers, terminals, system proxy...) | Pro |
The Pro tools need the mcp-write feature (llm-lens for akev_llm_conversation). Without
it, they return a tool error that names the feature; see licensing.
Safety
Agents read captured traffic, and a malicious server can put instructions into it. Akev therefore limits what an agent can do on your machine:
akev_activate_interceptor,akev_set_rulesandakev_clear_trafficare marked destructive, so clients that respect the hint ask you before running them.- The
electroninterceptor starts any executable with any arguments, so agents can't use it unless you opt in. Useakev start --mcp-allow-launch, orakev mcp --allow-launch("args": ["mcp", "--allow-launch"]) whenakev mcpruns its own instance. Without the opt-in, launch the app from the Connect page. - Secrets are redacted. Every tool that returns traffic shows
Authorization, cookies, API-key headers, tokens in query strings and similar values as[redacted]. A tool call can ask forreveal: trueonly when Akev was started withakev start --mcp-reveal-secrets(orakev mcp --reveal-secretsfor its own instance). - Files stay in known folders. Agents can only make Akev read and write files in the data
directory's
exportsfolder, plus folders you allow with `akev start --mcp-allow-path` (repeatable; `akev mcp --allow-path ` for its own instance). A HAR export to another path, or a rule that would serve a file from elsewhere (Map Local, a file body), is refused. Rules you make in the app aren't affected. - Only approve tool calls you asked for. Don't approve them because captured content asks you to.