You, the reader use a harness such ChatGPT, Claude, Cursor, Command Code, OpenCode etc everyday for coding, writing, editing, research and would like to in the words of the Hoff want to ‘Turbo Boost’ your harness by being able to do deeper research across topics.
So in order to do the above, I quote the Ghostbusters catch phrase ‘Who you gonna call?’ and humbly suggest reading the below document
TL;DR: The initial part of this document shows a suggested prefix which one can add to an appropriate prompt that incorporates multiple external search and fetch providers such as Exa, Octen and TinyFish.
Further down the page I provide a skill that you can drop into the skill folder of your harness and some snippets that you can add to the global instructions of your harness. I shall also provide some examples of prompts which invoke the skill and call one or more external search and fetch providers to better help you understand the value proposition of these external search and fetch providers
These approaches can be useful if you are looking to take advantage of specialized data that external search providers can bring to the table within your AI harness.
The mechanism described below an be useful for GTM intelligence , lead-generation, finance or investment research, competitive intelligence or market-mapping workflows, among others.
The assumption I’m making is that you are aware of the advantages of incorporating external search and fetch providers in a harness and utilizing them in queries compared to using the built in search/fetch mechanism which a harness provides
These answers from Grok and Perplexity can refresh your understanding
https://grok.com/share/bGVnYWN5LWNvcHk_8281139d-7bd3-438b-9958-9b2fe03c9f48
https://www.perplexity.ai/search/why-should-one-use-an-external-q8Pogu31SJyAYYJ.SWuVNw
The following assumes that you are using ChatGPT/Codex as a harness and have installed Exa (via its MCP), Octen via skills or MCP and Tinyfish as a plugin into that harness. I selected these providers because they provided a cheap entry point
This approach will also work if you are using other harnesses such as Claude Code/Claude Desktop, Cursor, Command Code, Opencode, Grok Build. I would recommend installing a provider at a time and then incorporating that provider into a query into the harness before moving to the next provider
Pro-tip for Exa. In order to be able to access all of its tools via its MCP server, don’t use its OAuth mechanism. If you use OAuth, then you can only access web_search_exa and web_fetch_exa which could be enough for many people Instead place the API key in the config.toml or equivalent file for other harnesses if you want access to all its tools
[mcp_servers.exa] url = "https://mcp.exa.ai/mcp?tools=web_search_exa,web_fetch_exa,web_search_advanced_exa,agent_run" http_headers = { "x-api-key" = "YOUR_EXA_API_KEY" }
A suggested prompt prefix to then utilise Exa, Octen and Tinyfish in a prompt is
<aside> 💡
Use Octen Broad Search for multi-angle discovery, passing the original query unchanged, and Octen Extract for its unique results.
For Exa, convert terse input into a semantically rich description of the ideal sources while preserving the original intent and constraints. Use regular Exa Search for independent semantic cross-checking, Exa Advanced Search only when precise domain, date, category, freshness, text, or subpage controls are needed, and Exa Fetch when highlights are insufficient or URLs are already known.
Before searching, inspect the currently callable Exa tools and the live agent_run schema. Use Exa Agent for multi-step research, enrichment, or structured list-building. Separately, whenever the task materially matches an available Exa Connect provider—even if I did not name that provider—invoke Agent with only the relevant supported dataSources. Explicitly request the provider-backed fields in outputSchema, inspect the completed run to confirm which providers actually contributed, and report their contributions. Do not infer availability from documentation alone or silently replace unavailable Connect data with web search.
Use TinyFish to verify JavaScript-heavy or interactive primary sources, or when direct extraction fails. Avoid duplicate discovery or fetching unless material verification requires it. Cite only successfully opened sources and briefly report provider contributions, failures, substitutions, and evidence gaps.
Query: [YOUR QUERY]
</aside>
This assigns distinct roles: