Skip to content

Use Cases

Jump to bottom
Svwergio_mel edited this page Nov 12, 2026 · 2 revisions

Use cases

What Chat Export is: the Chrome MV3 extension plus local MCP server at http://127.0.0.1:8765/mcp. It drives your logged-in provider tabs (Claude, ChatGPT, Grok, etc.) — list, export, ask — through the extension bridge. Not bundled Obsidian, Cursor, Claude Desktop, or local LLM stacks: those are optional external MCP clients you wire yourself. See Home for product scope.

At a glance

Use case One-liner
Second brain / Obsidian Pipe real provider answers into your vault via any MCP client that can write Obsidian; Chat Export fetches from live tabs.
Prompt / model bake-offs Same prompt across providers and models; compare quality, latency, and refusals with chat_ask and jobs polling.
Cursor / Claude Desktop research dump Offload “go ask the web” chores to MCP so agents hit providers through Chat Export instead of hallucinating or burning API keys.
Distillation datasets Harvest prompt→response pairs from live UIs for local fine-tune corpora — with eyes open on ToS and PII.
Fingerprint / session trail research Experiment: does the same provider answer differently under different accounts, locales, or session footprints?
Content farms (tongue-in-cheek) Multi-provider generate/export plumbing exists; we ship MIT AS-IS and do not endorse ToS abuse.

Tools you'll touch (conceptually): chat_ask; chats_list / chats_get / chats_export; async lanes via jobs_enqueue, jobs_status, jobs_list. Details: MCP tools and profiles.

One optional orchestrator example: local LLM via LangChain — pattern only; not required.

Second brain / Obsidian

You already have a vault, templates, and whatever ritual keeps your notes from rotting. Chat Export does not ship Obsidian integration. It ships access to the answers sitting in provider UIs you're already logged into.

Pattern: any local MCP client that can create or append Obsidian notes (Cursor agent, Claude Desktop, a small script, LangChain automation — your problem) calls Chat Export to chat_ask or chats_export, then writes markdown into the vault. The extension talks to the real tab; your vault tooling stays external.

Why bother: copy-paste from five chat sites is a second job. One MCP endpoint, multiple providers, exports land where your automation puts them.

Caveats: Advanced MCP mode and automation gates apply. Provider ToS and account risk: Legal & Two-Tier ToS on Home and LEGAL.md in the repos.

Prompt / model bake-offs

Research question, eval harness, or “which model actually answers this SQL dialect?” — same task, many destinations.

Pattern:

  1. Bind browsers / pick platforms (Supported platforms).
  2. Fire the same user message through chat_ask (or enqueue jobs_enqueue { "kind": "ask", … } and poll jobs_status when hosts are slow).
  3. chats_export or chats_get for side-by-side markdown/JSON.
  4. Score offline: quality rubric, latency from job metadata/logs, refusal detection.

Chat Export is the repeatable actuator on live UIs, not a leaderboard product. You own the scoring script.

Cursor / Claude Desktop as research dump

Your IDE agent is great at refactoring and terrible at pretending it read yesterday's Grok thread.

Pattern: point Cursor, Claude Desktop, or another MCP client at http://127.0.0.1:8765/mcp (installation). When the task is “summarize what Claude said about X” or “ask ChatGPT B and compare to A,” the agent calls chat_ask / list / export instead of inventing citations or routing everything through paid vendor APIs.

You keep your API keys for what they're for; web providers you already pay for (or use free tiers of) get exercised through your Chrome session.

Caveats: Tier 2 Advanced MCP Mode ToS before automation. Rate limits and bans are real — see Dashboard and security.

Distillation datasets

If you're building a local fine-tune or distillation corpus from your own interactions, Chat Export can mechanize collection: prompt in, full thread out via chats_export (markdown/JSON under server export paths).

This is not a license to scrape the internet. You are responsible for:

  • Each provider's Terms of Service and automation rules.
  • Account suspension risk (Tier 1 and Tier 2 ToS — Home, Installation).
  • Personal data in threads (yours and others'): redact before training; don't ship secrets to a dataset repo.

Chat Export provides plumbing; compliance and ethics are on you.

Fingerprint / “same provider, different trail” research

Half joke, half serious lab question: if you ask the same question on the same provider, do answers shift with different logged-in accounts, browser locales, VPN exit, or “fresh” vs aged sessions?

Chat Export lets you drive multiple bound tabs/sessions and compare outputs with the same MCP tools — reproducible runs, logged request_ids, exportable threads. We are not claiming conspiracy; we're offering experiment infrastructure on UIs you control.

Scope note: anti-detect browsers, Firefox support, and similar are future ideas — not this product today. Use separate Chrome profiles and honest experimentation.

Content farms (tongue-in-cheek)

Want a content farm? Congratulations, the pipes exist: multi-provider chat_ask, batch jobs_*, chats_export. Chat Export is MIT AS-IS plumbing between your browser and localhost MCP.

We do not endorse:

  • Violating provider ToS at scale.
  • Spam, impersonation, or bypassing safety systems because “it's local automation.”

Read LEGAL.md and Tier 2 opt-in before you automate. If your use case makes a compliance officer cry, that's a you problem — the wiki won't pretend otherwise.

Related

Clone this wiki locally