Prem MCP Server

Instructions

Implementation Reference

• src/index.ts:83-123 (handler)

  The handler function for the 'chat' tool. It constructs a PremChatRequest, calls the Prem API chat.completions.create, and returns the response as text content or error.

    const requestId = `chat-${Date.now()}-${Math.random().toString(36).substring(2, 7)}`;
    this.activeRequests.add(requestId);
    
    try {
      const chatRequest = {
        project_id: PROJECT_ID as string,
        messages: [{ role: "user", content: query }],
        ...(model && { model }),
        ...(system_prompt && { system_prompt }),
        ...(temperature && { temperature }),
        ...(max_tokens && { max_tokens }),
        ...(repository_ids && {
          repositories: {
            ids: repository_ids,
            similarity_threshold: similarity_threshold || 0.65,
            limit: limit || 3
          }
        })
      };
      
      const response = await this.client.chat.completions.create(chatRequest as any);
      const responseData = 'choices' in response ? response : { choices: [] };
  
      return {
        content: [{
          type: "text" as const,
          text: JSON.stringify(responseData, null, 2)
        }]
      };
    } catch (error) {
      return {
        content: [{
          type: "text" as const,
          text: `Chat error: ${error instanceof Error ? error.message : String(error)}`
        }],
        isError: true
      };
    } finally {
      this.activeRequests.delete(requestId);
    }
  }

• src/index.ts:69-124 (registration)

  Registration of the 'chat' tool with the MCP server using this.server.toolcall.

  this.server.tool(
    "chat",
    "Chat with Prem AI - supports chat completions with optional RAG capabilities.",
    {
      query: z.string().describe("The chat message to send"),
      system_prompt: z.string().optional().describe("Optional system prompt to guide the model's behavior"),
      model: z.string().optional().describe("Optional model to use for completion"),
      temperature: z.number().optional().describe("Optional temperature for response generation"),
      max_tokens: z.number().optional().describe("Optional maximum tokens to generate"),
      repository_ids: z.array(z.number()).optional().describe("Optional array of repository IDs for RAG"),
      similarity_threshold: z.number().optional().describe("Optional similarity threshold for RAG"),
      limit: z.number().optional().describe("Optional limit of context chunks for RAG")
    },
    async ({ query, system_prompt, model, temperature, max_tokens, repository_ids, similarity_threshold, limit }) => {
      const requestId = `chat-${Date.now()}-${Math.random().toString(36).substring(2, 7)}`;
      this.activeRequests.add(requestId);
      
      try {
        const chatRequest = {
          project_id: PROJECT_ID as string,
          messages: [{ role: "user", content: query }],
          ...(model && { model }),
          ...(system_prompt && { system_prompt }),
          ...(temperature && { temperature }),
          ...(max_tokens && { max_tokens }),
          ...(repository_ids && {
            repositories: {
              ids: repository_ids,
              similarity_threshold: similarity_threshold || 0.65,
              limit: limit || 3
            }
          })
        };
        
        const response = await this.client.chat.completions.create(chatRequest as any);
        const responseData = 'choices' in response ? response : { choices: [] };
  
        return {
          content: [{
            type: "text" as const,
            text: JSON.stringify(responseData, null, 2)
          }]
        };
      } catch (error) {
        return {
          content: [{
            type: "text" as const,
            text: `Chat error: ${error instanceof Error ? error.message : String(error)}`
          }],
          isError: true
        };
      } finally {
        this.activeRequests.delete(requestId);
      }
    }
  );

• src/index.ts:72-82 (schema)

  Zod schema for validating inputs to the 'chat' tool.

  {
    query: z.string().describe("The chat message to send"),
    system_prompt: z.string().optional().describe("Optional system prompt to guide the model's behavior"),
    model: z.string().optional().describe("Optional model to use for completion"),
    temperature: z.number().optional().describe("Optional temperature for response generation"),
    max_tokens: z.number().optional().describe("Optional maximum tokens to generate"),
    repository_ids: z.array(z.number()).optional().describe("Optional array of repository IDs for RAG"),
    similarity_threshold: z.number().optional().describe("Optional similarity threshold for RAG"),
    limit: z.number().optional().describe("Optional limit of context chunks for RAG")
  },
  async ({ query, system_prompt, model, temperature, max_tokens, repository_ids, similarity_threshold, limit }) => {

• src/types.ts:49-58 (schema)

  TypeScript interface defining the arguments for the chat tool.

  export interface ChatArgs {
    query: string;
    system_prompt?: string;
    model?: string;
    temperature?: number;
    max_tokens?: number;
    repository_ids?: number[];
    similarity_threshold?: number;
    limit?: number;
  } 

• src/types.ts:15-25 (helper)

  Type definition for the PremChatRequest used in the handler.

  export interface PremChatRequest {
    project_id: string;
    messages: PremMessage[];
    model?: string;
    system_prompt?: string;
    session_id?: string;
    temperature?: number;
    max_tokens?: number;
    stream?: boolean;
    repositories?: PremRepository;
  }