Use cases

Parlo turns APIs and cloud files into one thing every human, script, and agent already understands: a filesystem. Here are the problems that solves, end to end.

Give a coding agent any API, in 30 seconds, no MCP server

Problem: To let an agent use a new API, you normally find or write an MCP server, wire up its tools, and hope the schema fits in context. Every new service is another integration project.

Solution: Connect once. The agent uses commands it already knows. No SDK, no MCP server, no tool definitions.

1parlo connect stripe

Now any agent with a shell just explores:

1cat ~/parlo/stripe/README.md            # how this mount is laid out2ls ~/parlo/stripe/charges/3cat ~/parlo/stripe/charges/recent.jsonl  # latest items, one JSON object per line4grep -R "refunded" ~/parlo/stripe/charges/

The agent never imported a schema, registered a tool, or learned an SDK. Connect any API. Use it as files.

Debug a bug that spans three services

Problem: A billing bug touches GitHub, Linear, and Stripe. That means three SDKs, three auth flows, and a pile of tool descriptions burning context before any real work starts.

Solution: Every service is a folder. One grep crosses all of them.

1parlo mount2cd ~/parlo3 4grep -R "billing\|invoice\|subscription" github/issues linear/issues | head5cat github/issues/42.md6cat linear/issues/ENG-123.md7cat stripe/customers/cus_8x92.md8 9# Write the finding back to a cloud-backed drive.10mkdir -p main/reports11echo "Cause: subscription cache stale after Stripe webhook" > main/reports/billing-bug.md

Analyze a full dataset with SQL, no ETL pipeline

Problem: You want to run analytics over every charge, issue, or event. The API caps each request at one page, so you would normally build a paginating exporter and somewhere to put the rows.

Solution: Read all.jsonl. Parlo paginates the upstream API in the background and materializes the whole dataset as JSONL. It is just a file. Pipe it into any tool you already use.

1# First read may return "building" while Parlo paginates. Retry after ~30s.2cat ~/parlo/stripe/charges/all.jsonl | jq -s 'map(.amount) | add'3 4# Hand it to DuckDB, pandas, or anything that reads JSONL.5duckdb -c "SELECT status, count(*) FROM read_json_auto('~/parlo/stripe/charges/all.jsonl') GROUP BY 1"

No exporter to write, no warehouse to provision. The data lake builds itself.

Stay synced without re-pulling everything

Problem: Once you have the full export, new data keeps arriving. Re-downloading everything on every run is slow and wasteful.

Solution: Read the incremental export for only what changed since a given date.

1cat ~/parlo/stripe/charges/since-2026-05-01.jsonl | jq -s 'length'

Use all.jsonl for the full picture and since-{date}.jsonl for the delta. Both are materialized by Parlo, both are just files.

Search a giant repo without cloning it

Problem: You need to grep a huge codebase (v8, the Linux kernel) but cloning locally is slow, and recursive grep over a network mount is painful.

Solution: parlo rg runs ripgrep next to cached repo data, so heavy search happens server-side.

1parlo rg "TODO" vercel/next.js --path packages2parlo rg malloc -t cpp v8/v8 --stats

Run agents in CI and sandboxes without a mount

Problem: Your agent runs in a container or CI job where FUSE mounting is not available.

Solution: The same paths work through direct CLI ops, no mount required.

1parlo connections2parlo ls github/issues/3parlo cat linear/ENG-123.md4parlo write linear/ENG-123.md --from ./updated.md

Where to go next