@steipete/peekaboo-mcp
Version:
A macOS utility exposed via Node.js MCP server for advanced screen captures, image analysis, and window management
312 lines • 11.8 kB
JavaScript
import OpenAI from "openai";
export function parseAIProviders(aiProvidersEnv) {
if (!aiProvidersEnv || !aiProvidersEnv.trim()) {
return [];
}
return aiProvidersEnv
.split(/[,;]/) // Support both comma and semicolon separators
.map((p) => p.trim())
.filter(Boolean)
.map((provider) => {
const [providerName, model] = provider.split("/");
return {
provider: providerName?.trim() || "",
model: model?.trim() || "",
};
})
.filter((p) => p.provider && p.model);
}
export async function isProviderAvailable(provider, logger) {
const status = await getProviderStatus(provider, logger);
return status.available;
}
export async function getProviderStatus(provider, logger) {
try {
switch (provider.provider.toLowerCase()) {
case "ollama":
return await checkOllamaStatus(provider.model, logger);
case "openai":
return await checkOpenAIStatus(provider.model, logger);
case "anthropic":
return checkAnthropicStatus(provider.model);
default:
logger.warn({ provider: provider.provider }, "Unknown AI provider");
return {
available: false,
error: `Unknown provider: ${provider.provider}`,
};
}
}
catch (error) {
logger.error({ error, provider: provider.provider }, "Error checking provider status");
return {
available: false,
error: error instanceof Error ? error.message : "Unknown error",
};
}
}
async function checkOllamaStatus(model, logger) {
try {
const baseUrl = process.env.PEEKABOO_OLLAMA_BASE_URL || "http://localhost:11434";
// Check if server is reachable
const tagsResponse = await fetch(`${baseUrl}/api/tags`, {
signal: AbortSignal.timeout(3000), // 3 second timeout
});
if (!tagsResponse.ok) {
return {
available: false,
error: `Ollama server returned ${tagsResponse.status}`,
details: {
serverReachable: false,
},
};
}
const tagsData = await tagsResponse.json();
const availableModels = tagsData.models?.map((m) => m.name) || [];
// Check if the specific model is available
const modelAvailable = availableModels.some((m) => m === model || m.startsWith(model + ":") || model.startsWith(m.split(":")[0]));
if (!modelAvailable) {
return {
available: false,
error: `Model '${model}' not found. Available models: ${availableModels.join(", ") || "none"}`,
details: {
serverReachable: true,
modelAvailable: false,
modelList: availableModels,
},
};
}
return {
available: true,
details: {
serverReachable: true,
modelAvailable: true,
modelList: availableModels,
},
};
}
catch (error) {
logger.debug({ error }, "Ollama not available");
const errorMessage = error instanceof Error ? error.message : "Unknown error";
if (errorMessage.includes("fetch") || errorMessage.includes("timeout")) {
return {
available: false,
error: "Ollama server not reachable (not running or network issue)",
details: {
serverReachable: false,
},
};
}
return {
available: false,
error: errorMessage,
details: {
serverReachable: false,
},
};
}
}
async function checkOpenAIStatus(model, logger) {
const apiKey = process.env.OPENAI_API_KEY;
if (!apiKey) {
return {
available: false,
error: "OpenAI API key not configured (OPENAI_API_KEY environment variable missing)",
details: {
apiKeyPresent: false,
},
};
}
try {
// Test the API key by making a simple models list request
const openai = new OpenAI({
apiKey,
timeout: 3000, // 3 second timeout
});
const modelsResponse = await openai.models.list();
const availableModels = modelsResponse.data.map(m => m.id);
// Check if the specific model is available
const modelAvailable = availableModels.includes(model);
if (!modelAvailable) {
// For OpenAI, we'll be more lenient and just warn if model isn't in the list
// since the models list API might not include all available models
logger.debug({ model, availableCount: availableModels.length }, "Model not found in OpenAI models list, but this might be normal");
}
return {
available: true,
details: {
apiKeyPresent: true,
serverReachable: true,
modelAvailable: modelAvailable,
modelList: availableModels.slice(0, 10), // Limit to first 10 models for brevity
},
};
}
catch (error) {
logger.debug({ error }, "OpenAI API check failed");
const errorMessage = error instanceof Error ? error.message : "Unknown error";
if (errorMessage.includes("401") || errorMessage.includes("Unauthorized")) {
return {
available: false,
error: "Invalid OpenAI API key",
details: {
apiKeyPresent: true,
serverReachable: true,
},
};
}
if (errorMessage.includes("network") || errorMessage.includes("fetch")) {
return {
available: false,
error: "Cannot reach OpenAI API (network issue)",
details: {
apiKeyPresent: true,
serverReachable: false,
},
};
}
return {
available: false,
error: `OpenAI API error: ${errorMessage}`,
details: {
apiKeyPresent: true,
serverReachable: false,
},
};
}
}
function checkAnthropicStatus(_model) {
const apiKey = process.env.ANTHROPIC_API_KEY;
if (!apiKey) {
return {
available: false,
error: "Anthropic API key not configured (ANTHROPIC_API_KEY environment variable missing)",
details: {
apiKeyPresent: false,
},
};
}
return {
available: false,
error: "Anthropic support not yet implemented",
details: {
apiKeyPresent: true,
},
};
}
export async function analyzeImageWithProvider(provider, imagePath, imageBase64, question, logger) {
switch (provider.provider.toLowerCase()) {
case "ollama":
return await analyzeWithOllama(provider.model, imageBase64, question, logger);
case "openai":
return await analyzeWithOpenAI(provider.model, imageBase64, question, logger);
case "anthropic":
throw new Error("Anthropic support not yet implemented");
default:
throw new Error(`Unsupported AI provider: ${provider.provider}`);
}
}
async function analyzeWithOllama(model, imageBase64, question, logger) {
const baseUrl = process.env.PEEKABOO_OLLAMA_BASE_URL || "http://localhost:11434";
logger.debug({ model, baseUrl }, "Analyzing image with Ollama");
// Default to describing the image if no question is provided
const prompt = question.trim() || "Please describe what you see in this image.";
const response = await fetch(`${baseUrl}/api/generate`, {
method: "POST",
headers: {
"Content-Type": "application/json",
},
body: JSON.stringify({
model,
prompt,
images: [imageBase64],
stream: false,
}),
});
if (!response.ok) {
const errorText = await response.text();
logger.error({ status: response.status, error: errorText }, "Ollama API error");
throw new Error(`Ollama API error: ${response.status} - ${errorText}`);
}
const result = await response.json();
return result.response || "No response from Ollama";
}
async function analyzeWithOpenAI(model, imageBase64, question, logger) {
const apiKey = process.env.OPENAI_API_KEY;
if (!apiKey) {
throw new Error("OpenAI API key not configured");
}
logger.debug({ model }, "Analyzing image with OpenAI");
const openai = new OpenAI({ apiKey });
// Default to describing the image if no question is provided
const prompt = question.trim() || "Please describe what you see in this image.";
const response = await openai.chat.completions.create({
model: model || "gpt-4o",
messages: [
{
role: "user",
content: [
{ type: "text", text: prompt },
{
type: "image_url",
image_url: {
url: `data:image/jpeg;base64,${imageBase64}`,
},
},
],
},
],
max_tokens: 1000,
});
return response.choices[0]?.message?.content || "No response from OpenAI";
}
export function getDefaultModelForProvider(provider) {
switch (provider.toLowerCase()) {
case "ollama":
return "llava:latest";
case "openai":
return "gpt-4o";
case "anthropic":
return "claude-3-sonnet-20240229";
default:
return "unknown";
}
}
export async function determineProviderAndModel(providerConfig, configuredProviders, logger) {
const requestedProviderType = providerConfig?.type || "auto";
const requestedModelName = providerConfig?.model;
if (requestedProviderType !== "auto") {
// Find specific provider in configuration
const configuredProvider = configuredProviders.find((p) => p.provider.toLowerCase() === requestedProviderType.toLowerCase());
if (!configuredProvider) {
throw new Error(`Provider '${requestedProviderType}' is not enabled in server's PEEKABOO_AI_PROVIDERS configuration.`);
}
// Check if provider is available
const available = await isProviderAvailable(configuredProvider, logger);
if (!available) {
throw new Error(`Provider '${requestedProviderType}' is configured but not currently available.`);
}
const model = requestedModelName ||
configuredProvider.model ||
getDefaultModelForProvider(requestedProviderType);
return {
provider: requestedProviderType,
model,
};
}
// Auto mode - find first available provider
for (const configuredProvider of configuredProviders) {
const available = await isProviderAvailable(configuredProvider, logger);
if (available) {
const model = requestedModelName ||
configuredProvider.model ||
getDefaultModelForProvider(configuredProvider.provider);
return {
provider: configuredProvider.provider,
model,
};
}
}
return { provider: null, model: "" };
}
//# sourceMappingURL=ai-providers.js.map