mcp-octagon

Instructions

Implementation Reference

• src/index.ts:134-164 (handler)

  Handler function that invokes the Octagon 'octagon-scraper-agent' model via OpenAI-compatible client, processes the streaming chat completion response, and returns the formatted text content or error.

  async ({ prompt }: PromptParams) => {
    try {
      const response = await octagonClient.chat.completions.create({
        model: "octagon-scraper-agent",
        messages: [{ role: "user", content: prompt }],
        stream: true,
        metadata: { tool: "mcp" }
      });
  
      const result = await processStreamingResponse(response);
      return {
        content: [
          {
            type: "text",
            text: result,
          },
        ],
      };
    } catch (error) {
      console.error("Error calling Scraper agent:", error);
      return {
        isError: true,
        content: [
          {
            type: "text",
            text: `Error: Failed to process web scraping query. ${error}`,
          },
        ],
      };
    }
  }

• src/index.ts:132-133 (schema)

  Input schema defining the 'prompt' parameter for the tool.

    prompt: z.string().describe("Your natural language query or request for the agent"),
  },

• src/index.ts:128-165 (registration)

  MCP server tool registration for 'octagon-scraper-agent', including name, description, schema, and handler reference.

  server.tool(
    "octagon-scraper-agent",
    "[PUBLIC & PRIVATE MARKET INTELLIGENCE] Specialized agent for financial data extraction from investor websites. Capabilities: Extract structured financial data from investor relations websites, tables, and online financial sources. Best for: Gathering financial data from websites that don't have accessible APIs. Example queries: 'Extract all data fields from zillow.com/san-francisco-ca/', 'Extract all data fields from www.carvana.com/cars/'.",
    {
      prompt: z.string().describe("Your natural language query or request for the agent"),
    },
    async ({ prompt }: PromptParams) => {
      try {
        const response = await octagonClient.chat.completions.create({
          model: "octagon-scraper-agent",
          messages: [{ role: "user", content: prompt }],
          stream: true,
          metadata: { tool: "mcp" }
        });
  
        const result = await processStreamingResponse(response);
        return {
          content: [
            {
              type: "text",
              text: result,
            },
          ],
        };
      } catch (error) {
        console.error("Error calling Scraper agent:", error);
        return {
          isError: true,
          content: [
            {
              type: "text",
              text: `Error: Failed to process web scraping query. ${error}`,
            },
          ],
        };
      }
    }
  );

• src/index.ts:48-76 (helper)

  Shared helper function to process streaming responses from the Octagon API, aggregating content from chat completions or responses API chunks. Used by the tool handler.

  async function processStreamingResponse(stream: any): Promise<string> {
    let fullResponse = "";
    let citations: any[] = [];
  
    try {
      // Process the streaming response
      for await (const chunk of stream) {
        // For Chat Completions API
        if (chunk.choices && chunk.choices[0]?.delta?.content) {
          fullResponse += chunk.choices[0].delta.content;
  
          // Check for citations in the final chunk
          if (chunk.choices[0]?.finish_reason === "stop" && chunk.choices[0]?.citations) {
            citations = chunk.choices[0].citations;
          }
        }
  
        // For Responses API
        if (chunk.type === "response.output_text.delta") {
          fullResponse += chunk.text?.delta || "";
        }
      }
  
      return fullResponse;
    } catch (error) {
      console.error("Error processing streaming response:", error);
      throw error;
    }
  }