Support for commands

Author: ericdallo

CHANGELOG.md

@@ -3,6 +3,8 @@
 ## Unreleased
 
 - Fix openai api key read from config.
+- Support commands via `/`.
+- Support MCP prompts via commands.
 
 ## 0.11.2
 

docs/features.md

@@ -42,6 +42,10 @@ Here are the current supported contexts types:
 - `directory`: a directory in the workspace, server will read all file contexts and pass to LLM.
 - `repoMap`: a summary view of workspaces files and folders, server will calculate this and pass to LLM. Currently, the repo-map includes only the file paths in git.
 
+### Commands
+
+Eca supports commands that usually arer triggered via `/` in the chat, completing in the chat will show the known commands which include ECA commands and MCP prompts.
+
 ##  Completion
 
 Soon

docs/models.md

@@ -2,11 +2,11 @@
 
 ## Built-in providers and capabilities
 
-| model     | MCP / tools | thinking/reasioning | prompt caching | web_search |
-|-----------|-------------|---------------------|----------------|------------|
-| OpenAI    | √           | X                   | X              | √          |
-| Anthropic | √           | X                   | √              | √          |
-| Ollama    | √           | X                   | X              | X          |
+| model     | MCP / tools | MCP prompts | thinking/reasioning | prompt caching | web_search |
+|-----------|-------------|-------------|---------------------|----------------|------------|
+| OpenAI    | √           | √           | X                   | X              | √          |
+| Anthropic | √           | √           | X                   | √              | √          |
+| Ollama    | √           | √           | X                   | X              | X          |
 
 ### OpenAI
 

docs/protocol.md

@@ -626,14 +626,62 @@ interface ToolCallRejected {
 type ToolCallOrigin = 'mcp' | 'native';
 ```
 
+### Chat approve tool call (➡️)
+
+A client notification for server to approve a waiting tool call.
+This will execute the tool call and continue the LLM chat loop.
+
+_Notification:_
+
+* method: `chat/toolCallApprove`
+* params: `ChatToolCallApproveParams` defined as follows:
+
+```typescript
+interface ChatToolCallApproveParams {
+    /**
+     * The chat session identifier.
+     */
+    chatId: string;
+    
+    /**
+     * The tool call identifier to approve.
+     */
+    toolCallId: string; 
+}
+```
+
+### Chat reject tool call (➡️)
+
+A client notification for server to reject a waiting tool call.
+This will not execute the tool call and return to the LLM chat loop.
+
+_Notification:_
+
+* method: `chat/toolCallReject`
+* params: `ChatToolCallRejectParams` defined as follows:
+
+```typescript
+interface ChatToolCallRejectParams {
+    /**
+     * The chat session identifier.
+     */
+    chatId: string;
+    
+    /**
+     * The tool call identifier to reject.
+     */
+    toolCallId: string; 
+}
+```
+
 ### Chat Query Context (↩️)
 
-A request sent from client to server, querying for the available context to user add to prompt calls.
+A request sent from client to server, querying for all the available contexts for user add to prompt calls.
 
 _Request:_ 
 
 * method: `chat/queryContext`
-* params: `ChatQueryParams` defined as follows:
+* params: `ChatQueryContextParams` defined as follows:
 
 ```typescript
 interface ChatQueryContextParams {
@@ -670,51 +718,65 @@ interface ChatQueryContextResponse {
 }
 ```
 
-### Chat approve tool call (➡️)
+### Chat Query Commands (↩️)
 
-A client notification for server to approve a waiting tool call.
-This will execute the tool call and continue the LLM chat loop.
+A request sent from client to server, querying for all the available commands for user to call.
+Commands are multiple possible actions like MCP prompts, doctor, costs. Usually the 
+UX follows `/<command>` to spawn a command.
 
-_Notification:_
+_Request:_ 
 
-* method: `chat/toolCallApprove`
-* params: `ChatToolCallApproveParams` defined as follows:
+* method: `chat/queryCommands`
+* params: `ChatQueryCommandsParams` defined as follows:
 
 ```typescript
-interface ChatToolCallApproveParams {
+interface ChatQueryCommandsParams {
     /**
      * The chat session identifier.
      */
-    chatId: string;
-    
+    chatId?: string;
+
     /**
-     * The tool call identifier to approve.
+     * The query to filter results, blank string returns all available commands.
      */
-    toolCallId: string; 
+    query: string;
 }
 ```
 
-### Chat reject tool call (➡️)
+_Response:_
 
-A client notification for server to reject a waiting tool call.
-This will not execute the tool call and return to the LLM chat loop.
+```typescript
+interface ChatQueryCommandsResponse {
+    /**
+     * The chat session identifier.
+     */
+    chatId?: string;
 
-_Notification:_
+    /**
+     * The returned available Commands.
+     */
+    commands: ChatCommand[];
+}
 
-* method: `chat/toolCallReject`
-* params: `ChatToolCallRejectParams` defined as follows:
+interface ChatCommand {
+    /**
+     * The name of the command.
+     */
+    name: string;
 
-```typescript
-interface ChatToolCallRejectParams {
     /**
-     * The chat session identifier.
+     * The description of the command.
      */
-    chatId: string;
+    description: string;
 
     /**
-     * The tool call identifier to reject.
+     * The arguments of the command.
      */
-    toolCallId: string; 
+    arguments: [{
+       name: string;
+       description?: string;
+       required: boolean; 
+    }];
 }
 ```