Blink Connectors

Getting Started

# List available providers
blink connector providers

# Check connection status
blink connector status slack

# Execute an API call
blink connector exec slack /conversations GET

MCP Tools

Prerequisites

Initial OAuth connection must be done in the browser at blink.new/settings?tab=connectors. Users authorize each provider once, then CLI/MCP/SDK can make API calls.

SDK Usage

// Check status
const status = await blink.connectors.status('slack')

// GET request
const channels = await blink.connectors.execute('slack', {
  method: '/conversations',
  http_method: 'GET',
  params: { types: 'public_channel,private_channel' }
})

// POST request
await blink.connectors.execute('slack', {
  method: '/chat/postMessage',
  http_method: 'POST',
  params: { channel: 'C1234', text: 'Hello from Blink!' }
})

Available Providers

Plus 20+ more — run blink connector providers for the full list.

CLI Examples

# Slack: list channels
blink connector exec slack /conversations GET

# Notion: search pages
blink connector exec notion /search POST '{"query":"meeting notes"}'

# Google Sheets: read data
blink connector exec google_sheets "/spreadsheets/SHEET_ID/values/Sheet1!A1:D10" GET

# HubSpot: list contacts
blink connector exec hubspot /contacts GET '{"limit":50}'

# Salesforce: SOQL query
blink connector exec salesforce /query GET '{"q":"SELECT Id,Name FROM Account LIMIT 10"}'

CLI syntax: blink connector exec <provider> <endpoint> [method] [params]

Method Path Rules

1. Paths start with / — /conversations not conversations
2. Dynamic IDs go in the path — /channels/${channelId}/messages
3. Provider IDs use underscores — google_drive, google_calendar, google_sheets
4. GET params become query parameters; POST params become JSON body
5. Invalid JSON in CLI params hard-fails (exit 1) — bad JSON used to silently send {}. If you see Invalid JSON for params: ..., fix the quoting (shell-escape inner double quotes or use --input @file.json).

Google Ads (composio_googleads) — reading data

Every path must start with the API version, and the customer id goes in the PATH, never in the body. There is no top-level /search resource — /search, /googleads/search and /searchStream all 404.

# Find the customer id
blink connector exec composio_googleads /v23/customers:listAccessibleCustomers GET

# Run a report (GAQL)
blink connector exec composio_googleads /v23/customers/CUSTOMER_ID/googleAds:search POST \
  '{"query":"SELECT campaign.name, metrics.clicks FROM campaign WHERE segments.date DURING LAST_14_DAYS"}'

googleAds:searchStream works the same way but returns its payload — errors included — wrapped in an array.

Quota (RESOURCE_EXHAUSTED, scope DEVELOPER)

A 429 with rateScope: DEVELOPER is a cap on Composio's shared developer token — pooled across tenants — not on the user's Google Ads account. Nothing the user does to their account fixes it. Honour the retryDelay (often 10h+) and surface it to the user rather than retrying into it.

rateScope: ACCOUNT means the opposite: that one is the user's own account rate limit, and the delay is usually short.

Google Ads (composio_googleads) — writing campaigns

The old falsy-stripping workarounds are retired. Composio's proxy used to drop false, 0, {} and [] from POST bodies (#3324). That was fixed upstream on 2026-07-24 and re-verified against the live proxy on 2026-07-29. Send falsy values normally. Do not build portfolio bidding strategies just to avoid manualCpc: {}, and do not omit a network setting you want off — omitting it means "use account defaults", which is not the same as false.

1. Dry-run first with partialFailure: true + validateOnly: true. Google returns 200 with a partialFailureError carrying the full fieldPathElements list, so you see every bad field at once instead of one error per round trip. Flip validateOnly to false only once the dry-run is clean.

2. Empty proto markers work inline. { "manualCpc": {} } satisfies campaign_bidding_strategy — verified. Portfolio strategies are still the right choice when you actually want shared bidding across campaigns, but they're no longer a workaround.

3. finalUrls goes at the Asset top level, NOT inside sitelinkAsset. This one still holds, but it's plain Google schema, not a proxy quirk: SitelinkAsset has no finalUrls field, so nesting it returns Invalid JSON payload ... Unknown name "finalUrls".

4. v23 requires containsEuPoliticalAdvertising on campaign create. Use the string enum: "containsEuPoliticalAdvertising": "DOES_NOT_CONTAIN_EU_POLITICAL_ADVERTISING". Omitting it returns contains_eu_political_advertising: REQUIRED.

5. networkSettings booleans are real booleans. Set them true or false as you mean them. Note Google rejects targetGoogleSearch: false on a Search campaign with OPERATION_NOT_PERMITTED_FOR_CONTEXT — that's Google's rule, not a stripped field.

Debugging recipe

# Dry-run and read the full error list
blink connector exec composio_googleads /v23/customers/CUSTOMER_ID/campaigns:mutate POST \
  '{"operations":[...],"partialFailure":true,"validateOnly":true}'

# Then: read partialFailureError.details[].errors[].location.fieldPathElements
# to see exactly which field Google rejected, fix it, re-run with validateOnly:false

Prefer connector exec (the proxy) over connector tool-execute (the native tool catalog) for Google Ads. The native path collapses Google's error into an opaque Error executing the tool GOOGLEADS_* with no field detail.

Error Codes

Token refresh is handled automatically by Blink backend.