I built a research bridge that lets my AI assistant search LinkedIn through my signed-in account. Once I have approved access, the agent can run separate research tasks without asking me to repeat the same setup each time.
The aim was to make the information available through my accounts easier to use in research, content planning and commercial work. I wanted to ask for research, let the agent collect relevant information and work with the results in the same conversation.
Building the research bridge
Working with Codex, I adapted the open-source browser automation project OpenCLI into Personal Research Bridge. The setup runs on my Windows computer and connects my chosen Chrome profile to local AI assistants.
It also includes an MCP connection, which gives compatible AI assistants a common way to use the tool. This makes the research capability available across my local projects without rebuilding the integration for every new task.
I set the priorities: useful access, free research tools where available and minimal repetitive setup. Codex handled source review, implementation and testing. I also used additional AI agents to review the changes, investigate problems and check the fixes.
Approving sources for repeated research
One important decision was to separate my approval from the access needed for an individual task. I choose the account and approve the source. The bridge then opens temporary access when research is requested and closes it when the task finishes. My approval stays in place for the next request, and I can disable access when needed.
That makes repeated research much more practical. An agent can collect information from an approved source while I stay focused on the question I want answered.
Testing repeated use
Getting that repeated use to work took more attention than the initial installation. During live testing, a search would succeed and later requests would fail. The agents traced the problem to how the browser handled navigation. An existing OpenCLI correction addressed the issue, so we adapted that fix and checked the workflow again.
A working LinkedIn demonstration
The final demonstration used two consecutive LinkedIn job searches: “AI marketing” and “AI strategy.” Each returned a job record with a direct source link. Both completed without another approval click, and temporary access ended after each search.
This established a repeatable connection between my AI workflow and a signed-in research source. The assistant could collect structured information and bring it back with links I could follow for further investigation.
Connecting research to commercial questions
For my work, that opens several useful directions. Job listings can provide starting points for understanding how companies describe AI responsibilities, which skills they seek and where commercial roles are changing. With a focused research brief and a broader set of results, an agent could help compare those requirements and develop topics for articles, services or further market research.
The same approach could support competitive research as more sources and workflows are connected and tested. An agent could gather relevant material, preserve the source links and prepare it for comparison. I would then spend more of my attention interpreting the findings and deciding what to do with them.
Developing repeatable business briefs
Building the bridge also demonstrated a useful way to work with AI agents: define the business need, adapt existing software, use separate agents to review the work and test the complete process in the environment where it will actually run.
The next opportunity is to connect this research capability to repeatable business briefs. Hiring signals, competitor developments and content opportunities could each have their own questions, sources and output format. That would give the agents a clearer purpose and make each research run easier to turn into a useful decision.