Examples and workflows

Every example is a plain-language prompt you can paste into your assistant today, followed by what happens. No SDK, no ceremony: the model does the calling.

Memory across sessions

Save our decisions so far to Pastepile as a decision record titled "auth: decision record".

The assistant calls `pastepile_save` and keeps the URL in context. Tomorrow, in a fresh session:

Search my pastes for the auth decision record and continue where we left off.

That is `pastepile_search` with scope mine, then `pastepile_get`. The whole loop is three tool calls and zero file management.

Runbooks as tool calls

Check my pastes for how we deploy the worker, then walk me through it.

Keep runbooks, snippets, and prompts as pastes with prefixed titles (infra: deploy worker). Tribal knowledge becomes something any assistant can look up mid-task.

Handoff between agents

Save your analysis to Pastepile and print the slug; the reviewer agent will pick it up.

One agent saves, another (any MCP client, any machine, even the CLI) retrieves by slug. Pastepile acts as the shared blackboard, and the URL is the interface.

Code review notes that outlive the chat

Summarize this review as a checklist and save it to Pastepile so I can send the link to the team.

Any saved paste is a URL teammates open with no account and no client. Unlisted visibility means only people with the link see it.

From CI, without MCP

The same memory is writable from places MCP does not reach: the GitHub Action pastes failing test logs from CI and can comment the URL on the PR, and your assistant reads it back with pastepile_get. One knowledge store, many writers.

Transform what you retrieve

Retrieved a JWT, minified JSON, or Base64 blob? The Developer Toolbox decodes, formats, and converts in the browser, and the paste viewer's Content Intelligence suggests the right tool automatically. For in-chat transforms, assistants handle decoding themselves; the Toolbox is the human-facing complement.