Guides

API connections

Connect documented APIs as folders of readable, editable resources.

Parlo mounts documented APIs as folders.

bash

3 lines

1ls ~/parlo/github/issues/2cat ~/parlo/linear/ENG-123.md3vim ~/parlo/github/issues/42.md

API resources become markdown files. Reading a file calls the API. Saving a file can update the remote resource.

Connect from a provider or intent

You can choose a provider:

GitHub
Stripe
Linear

Or describe what you want:

API for wrestling matches
weather alerts API
restaurant health inspection API

Parlo searches for real API documentation, asks what you want mounted, generates an adapter, and tests it before saving.

Paste a URL (fastest path)

If you already have a spec URL, paste it directly on the connections page. Parlo auto-detects the format (OpenAPI, GraphQL, or MCP server), compiles the adapter, and connects without wizard steps. The same works from the CLI:

bash

4 lines

1parlo connect --spec https://api.example.com/openapi.json2parlo connect --openapi https://api.example.com/openapi.json3parlo connect --graphql https://api.example.com/graphql4parlo connect --mcp https://example.com/mcp

If the spec requires auth and no token is provided, Parlo returns a structured auth-required response instead of failing silently.

How generation works

1Find official docs, specs, schemas, or examples.
2Ask which resources should appear as files.
3Generate a filesystem adapter for that scope.
4Ask for your API key only when ready to test.
5Run the adapter in a sandbox.
6Save the connection if the test passes.

Generated adapters

A generated adapter translates file operations into API calls.

File action	API behavior
`ls`	list resources
`cat`	fetch and format a resource
save / `write`	create or update
`rm`	delete, close, or clean up

Adapters are generated from documented APIs. If Parlo cannot find enough information, it asks for a docs URL or another provider instead of guessing.

Credentials

The wizard does not ask for an API key up front. It asks only when the adapter is ready to run a sandbox test. Saved credentials are encrypted and injected by Parlo at runtime. Adapter code never handles secrets directly.

Reliable by design

Adapters implement a small filesystem contract, so every connection behaves the same way:

Operation	Purpose
`list`	show files and folders
`read`	return one resource as markdown or bytes
`stat`	return metadata like size and content type
`write`	create or update a resource
`parse_webhook`	invalidate cached paths when remote data changes

Two things keep generated connections dependable for long-running use, agents, and background sync:

Documented APIs only. Parlo generates from official docs, specs, and schemas. It does not scrape websites or depend on page markup. If it cannot find enough information, it asks for a docs URL instead of guessing.
Sandbox-tested before saving. A connection is only saved after a sandbox run proves authentication works and the mount returns a real listing. If the test fails, the wizard revises and retries.

Semantic discovery

toolSearch (MCP) and /api/connections/discover find tools by intent using vector similarity via Cloudflare Vectorize and Workers AI embeddings. Search for "billing API" or "customer records" and get the closest match, not just a name substring. Tools are auto-embedded when connections are saved. Falls back to text matching when embeddings are unavailable.

Example shapes

A project management API might look like:

bash

4 lines

1~/parlo/linear/2  README.md3  ENG-123-fix-login.md4  ENG-124-update-billing.md

A search API might look like:

bash

4 lines

1~/parlo/search/rust-vs-go/2  README.md3  1-rust-vs-go-performance.md4  2-when-to-use-rust.md

← Previous

Filesystem model

Cloud drives