const apiEndpointPrefix = atom.config.get('pulsar-edit-mcp-server.apiEndpointPrefix');
const apiKey = atom.config.get('pulsar-edit-mcp-server.apiKey');
const hljs = require('highlight.js');
var mcpTools = null;
var chatDisplay = null;
var marked = null;
var DOMPurify = null;
var currentModel = null;
var cachedModels = null;
var chatObj = null;
var llmContextHistory = [{
role: "system",
content: "You are a helpful coding assistant with access to the user's Pulsar editor IDE."
}];
function updateChatHistory(sender, markdownText) {
if (!chatDisplay) return console.error('Chat display missing');
const rawHtml = marked.parse(markdownText, { breaks: true });
const safeHtml = DOMPurify.sanitize(rawHtml);
const msg = document.createElement('div');
msg.classList.add('message', sender.toLowerCase());
msg.innerHTML = safeHtml;
chatDisplay.appendChild(msg);
msg.querySelectorAll('pre code').forEach(block => hljs.highlightElement(block));
msg.scrollIntoView({ block: 'end' });
}
async function getMcpTools(mcpClient) {
if (mcpTools == null) {
const { tools } = await mcpClient.listTools();
console.log("Fetching available tools");
var openAiTools = tools.map(t => ({
type: "function",
function: {
name: t.name,
description: t.description ?? "",
parameters: t.inputSchema
}
}));
} else {
console.log("Found available tools");
}
return openAiTools;
}
async function chatToLLM(message, mcpClient) {
var newContext = {
role: "user",
content: message
};
updateLlmContextHistory(newContext);
await callLLM(mcpClient);
}
function addToolResultToHistory(tc, toolResult) {
const newContext = {
role: "tool",
name: tc.function.name,
content: JSON.stringify(toolResult.content),
tool_call_id: tc.id
};
updateLlmContextHistory(newContext);
}
async function callLLM(mcpClient) {
let returnMessage = null;
const availableTools = await getMcpTools(mcpClient);
const requestData = {
model: currentModel,
messages: llmContextHistory,
tools: availableTools,
tool_choice: "auto",
max_tokens: 1000,
};
try {
const response = await fetch(apiEndpointPrefix + '/v1/chat/completions', {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'Authorization': `Bearer ${apiKey}`
},
body: JSON.stringify(requestData),
});
if (!response.ok) {
throw new Error(`Error: ${response.statusText}`);
}
const data = await response.json();
console.log("RECEIVED: " + JSON.stringify(data));
updateLlmContextHistory(
data.choices[0].message
);
// Handle Tool Calls
const toolCalls = data.choices[0].message.tool_calls ?? [];
for (const tc of toolCalls) {
const args = JSON.parse(tc.function.arguments || "{}");
const toolResult = await mcpClient.callTool({ name: tc.function.name, arguments: args });
addToolResultToHistory(tc, toolResult);
}
if (toolCalls.length > 0) {
return await callLLM(mcpClient);
}
// Update chat area with message response after all tools have been executed
returnMessage = data.choices?.[0]?.message?.content?.trim?.() || "";
if (returnMessage) {
updateChatHistory('Assistant', returnMessage);
chatObj.thinkingOnOff(false);
}
} catch (error) {
console.error('Error calling LLM API:', error);
throw error;
}
}
async function fetchModels() {
if (cachedModels) {
console.log("RETURNING CACHED MODEL LIST");
return cachedModels;
}
console.log("FETCHING MODEL LIST");
const res = await fetch(apiEndpointPrefix + '/v1/models', {
headers: { 'Authorization': `Bearer ${apiKey}` }
});
if (!res.ok) throw new Error(`Model list failed: ${res.statusText}`);
const data = await res.json();
cachedModels = data.data?.map(m => m.id) ?? [];
console.log("Available models:", cachedModels);
return cachedModels;
}
async function handleSendMessage(inChatObj, inChatDisplay, inMarked, inDOMPurify, message, inModel, mcpClient) {
console.log('Message from UI:', message);
chatObj = inChatObj;
chatObj.thinkingOnOff(true);
currentModel = inModel;
chatDisplay = inChatDisplay;
marked = inMarked;
DOMPurify = inDOMPurify;
updateChatHistory('User', message);
chatToLLM(message, mcpClient);
}
function updateLlmContextHistory(inMessage) {
// llama.cpp server does not like null content
if (!inMessage.hasOwnProperty('content')) {
inMessage['content'] = "";
}
if (inMessage['content'] == null) {
inMessage['content'] = "";
}
llmContextHistory.push(inMessage);
}
function clearContextHistory() {
llmContextHistory.length = 0;
}
module.exports = {
updateChatHistory,
callLLM,
fetchModels,
handleSendMessage,
clearContextHistory
};
