MCP Server

@keq-request/cli includes a built-in Model Context Protocol (MCP) server that exposes your project's API directory as tools callable by AI assistants (Claude, Cursor, etc.).

Starting

keq mcp [options]

Option	Description
-c --config <config_file_path>	Path to the configuration file
--tolerant	Tolerant mode: ignores invalid Swagger/OpenAPI document structure
--debug	Print debug information

The MCP Server communicates via the stdio transport protocol, making it suitable for launching as a subprocess of an AI editor.

Configuring AI Editors

tip

It's recommended to configure the MCP Server in the project directory rather than globally, so each project can independently manage its own API context.

Claude Code

Add to .mcp.json in your project root:

{
  "mcpServers": {
    "keq": {
      "command": "npx",
      "args": ["@keq-request/cli", "mcp"]
    }
  }
}

Cursor

Add to .cursor/mcp.json in your project:

{
  "mcpServers": {
    "keq": {
      "command": "npx",
      "args": ["keq", "mcp"]
    }
  }
}

VS Code

Add to .vscode/mcp.json in your project:

{
  "servers": {
    "keq": {
      "command": "npx",
      "args": ["keq", "mcp"]
    }
  }
}

Available Tools

The MCP Server registers the following tools for AI assistants to call:

search_apis

Semantic search for API endpoints using natural language, supporting cross-language matching in Chinese and English.

Parameter	Type	Required	Description
query	string	Yes	Natural language search query
module	string[]	No	Filter by module name
limit	number	No	Maximum number of results (default: 10)

get_api_detail

Get the complete definition of a specified API, including request parameters, request body, and response body JSON Schema.

Parameter	Type	Required	Description
module	string	Yes	Module name
method	string	Yes	HTTP method (GET, POST, PUT, DELETE, etc.)
pathname	string	Yes	API path (e.g., /api/v1/users)

list_modules

List all API modules configured in .keqrc.

No parameters required.

list_apis

List structured information for all API endpoints (module, method, path, operationId, summary).

Parameter	Type	Required	Description
module	string[]	No	Filter by module name
method	string	No	Filter by HTTP method
pathname	string	No	Filter by path (supports glob wildcards)
includes	string[]	No	Content to include, options: operations, components (default: ['operations'])

build_apis

Generate type-safe TypeScript client code for specified API endpoints. Files are written to the output directory configured in .keqrc.

Parameter	Type	Required	Description
module	string[]	No	Modules to generate code for
method	string	No	Filter by HTTP method
pathname	string	No	Filter by path (supports glob wildcards, e.g., /api/v1/users/**)
fresh	boolean	No	Clear output directory before building (default: false)

list_generated_files

List generated TypeScript client files. Optionally list only invalid/outdated files.

Parameter	Type	Required	Description
invalid	boolean	No	When true, only lists outdated files not in current build output (default: false)

get_filter_rules

View the current .keqfilter file contents. Filter rules control which APIs generate code.

No parameters required.

add_filter_rule

Add a filter rule to .keqfilter. deny rules exclude APIs from code generation, allow rules include them.

Parameter	Type	Required	Description
mode	string	Yes	Rule mode: deny or allow
module	string	No	Module name pattern (supports glob, default: *)
method	string	No	HTTP method pattern (default: *)
pathname	string	Yes	Path pattern (supports glob)
build	boolean	No	Whether to automatically rebuild after adding the rule (default: false)

remove_filter_rule

Remove a specified filter rule from .keqfilter.

Parameter	Type	Required	Description
mode	string	Yes	Block the rule belongs to: deny or allow
module	string	Yes	Exact match module name pattern
method	string	Yes	Exact match method pattern
pathname	string	Yes	Exact match path pattern
build	boolean	No	Whether to automatically rebuild after removing the rule (default: false)

Auto-Reload

The MCP Server monitors changes to .keqrc and .keqfilter files. When these files are modified, the server automatically recompiles the API index without requiring a manual restart.

Process Lifecycle

The MCP Server runs as a stdio subprocess, launched on demand by the AI editor. To prevent multiple editor windows from accumulating long-lived processes and exhausting system memory, the server automatically exits in any of the following cases:

• The client disconnects (the editor closes or drops the MCP connection)
• The parent process exits (the editor process disappears)
• A termination signal is received (SIGINT / SIGTERM / SIGHUP)
• It stays idle for over 30 minutes without any tool call

No manual intervention is needed after it exits — the AI editor will automatically restart the server on next use.

tip

The semantic search model (used by search_apis) is lazily loaded, only loaded on the first search_apis call. Using only list_apis, list_modules, get_api_detail, and other tools will not consume model memory. As a result, after a process restart, only the first semantic search incurs a one-time model loading delay.