@hasna/connectors weekly downloads statistics

@hasna/connectors

Open-source connector platform for enabling, authenticating, and running API connectors from one package.

Install

What It Is

@hasna/connectors is a single-product runtime:

• connectors stay inside this repo and package • projects enable the connectors they want through .connectors/manifest.json • credentials live in the connectors data home (resolved by the in-package resolver; ~/.hasna/connectors/ until the XDG data home is adopted) • CLI, MCP, REST, and the local API/OAuth server all use the same connector registry

Local or remote

Connectors runs locally by default, with nothing to configure: the CLI, the MCP server and the SDK run the connector engine in-process and keep connections in the local store.

To use a server instead (one you run yourself, or any service that speaks the /v1 API), point the client at it:

Settings resolve in this order: the --api-url / --api-key-stdin flags, then CONNECTORS_API_URL / CONNECTORS_API_KEY, then /config.json (mode 0600, written by connectors login), then local mode. A saved key is only sent to the server it was saved for, a server URL without a key is an error rather than a silent fall back to local mode, and plain http:// is refused for anything but loopback unless CONNECTORS_ALLOW_INSECURE_HTTP=1.

Service connectors work the same way in both modes:

The shipped service connectors are anthropic, brandsight, cloudflare, cursor, godaddy, openai, opencode, stable and youtrust (Youtrust, formerly Yousign: signature requests, documents, recipients, audit trails and webhooks; see docs/providers/youtrust.md). A Youtrust connection picks its environment with a plain field:

Secret connection fields are never accepted as arguments (--field refuses them); use --secret-env key=ENV_NAME, --secrets-stdin or the hidden prompt. Any operation that is not a read (write, billable, destructive, credential) needs --confirm. A credential result, such as a newly minted key, is shown once and never stored.

CLI

Typical flow:

Key commands:

• connectors install writes project enablement to .connectors/manifest.json • connectors list and connectors search browse the shared catalog • connectors docs reads connector docs from the internal registry • connectors auth stores credentials in the connectors data home (resolved by the in-package resolver) • connectors auth --key-stdin reads an API key or bearer token from stdin, and --key-env reads it from the named environment variable. connectors setup and connectors llm set take the same two flags. Do not pass a key as a command-line value: it lands in shell history and the process list. • connectors run executes connector commands from the one-product runtime • connectors status and connectors doctor verify setup and auth state

Compact Output Defaults

CLI output is compact by default so agent terminals do not fill with huge catalogs, README sections, credential status tables, or API records.

• connectors list shows a compact page. Use --cursor or --offset for the next page, --limit to change page size, --verbose for wider rows, or --all --verbose for the full human catalog. --json now returns a minified 20-row envelope with truthful total, limit, cursor, next_cursor, and has_more; use --json --verbose for full fields within that bounded page, or --json --full / --json --all for the legacy bare array. Bounded JSON limits above 100 are refused with full-output guidance. • connectors docs shows a summary. Use --verbose for full parsed sections, --essential for auth/env vars only, --raw for raw markdown, or --json for structured docs. • connectors status, connectors doctor, connectors jobs list, connectors workflows list, connectors whoami, and dry-run install output page or truncate human rows by default. Use --verbose, show/inspect, or --json when full detail is needed. • connectors run preserves JSON-style command output, but truncates very large human stdout. Use connectors run --verbose ... for full stdout.

MCP Server

MCP tools also use compact defaults. list_connectors, list_jobs, list_workflows, and list_agents return paged objects with count, total, limit, cursor, nextCursor, and hasMore; pass cursor to continue, limit to change page size, and verbose: true for full records. list_connectors page receipts bypass optional probabilistic output stripping, so continuation metadata cannot be removed. connector_docs and operation discovery return summaries unless verbose: true or a specific detail path is requested.

Service-connector tools run through the same backend as the CLI (local, or the server from CONNECTORS_API_URL / connectors login): connectors_service_list, connectors_service_describe, connections_list, connections_check, connectors_call (pass confirm: true for anything that is not a read) and runs_list. No tool accepts a secret value; connections are created with the CLI or the API, so raw keys never pass through an agent's tool arguments. When connectors_call runs a credential operation (one that mints a key), the secret values in its result (key, token, secret, api_key, access_token, refresh_token, client_secret, value, at any depth) come back as [redacted: pass reveal_secret=true to return it] unless the call sets reveal_secret: true. The CLI's connectors call still prints the value, once, with a warning.

HTTP mode

Health: GET http://127.0.0.1:8808/health. MCP is also mounted on connectors-serve at /mcp.

Run your own server

connectors-serve serves the /v1 API that the CLI, MCP server and SDK use in remote mode.…