Menu

Mattering Automations Jina AI integration

Connect Jina AI to Mattering Automations: Reader, web search, embeddings, reranking, classification, free segmentation, and batch embedding jobs as workflow steps, authenticated with a Jina API key.

The Mattering Automations Jina AI integration adds Jina's search foundation to your automation canvas: turn any URL into LLM-ready markdown with Reader, search the web as structured JSON, create embeddings, rerank documents, classify text or images, train classifiers, segment long text for free, and run batch embedding jobs for large corpora.

Setup takes a couple of minutes: get an API key from jina.ai, attach it as a credential in Mattering, and test the connection. One key works across all three Jina hosts the integration uses (api.jina.ai, r.jina.ai, s.jina.ai). This guide walks through each step and ends with the full node reference.

Nothing in this integration is destructive: every node is a pure read or a compute call, and the strongest side effects anywhere are creating a batch job and training a classifier. The thing to manage is token spend, and every response carries a usage figure for exactly that.

Before you start

  • A Jina AI account with an API key (jina.ai shows the key and its remaining token balance).
  • Token balance for the metered nodes. Jina: Segment Text is free; Reader, Search, embeddings, reranking, and classification meter by tokens processed.
  • A Mattering workspace with access to the Automations app.

Get a Jina AI API key

Mattering authenticates to Jina with an API key, sent as a bearer credential on every request to all three hosts. There are no scopes to choose; one key covers Reader, Search, and the core API.

Sign in at jina.ai and copy your API key from the dashboard.
Check the key's token balance while you are there: metered nodes draw from it.
Jina allows some endpoints to be called without any key at low rate limits, but the integration always authenticates, which raises rate limits and makes usage attributable to your account.

Connect Jina AI to Mattering

In Mattering Automations, open Connectors and click Add Connector.
Choose Jina AI from the list.
Click Create new credential, paste your API key, and save.
The credential is stored encrypted in Google Secret Manager and never appears in workflow logs.
Click Test Connection: Mattering calls Jina's Segmenter, which costs zero tokens yet still rejects invalid keys, so the test is a true auth check that spends nothing.
Click Connect. All 11 Jina nodes appear in your canvas palette.

Verify it works

After connecting, open any automation canvas and check the palette: a Jina section should list the query and action nodes. Drop Jina: Segment Text onto the canvas and run a test step with any paragraph: it is free, and a successful response returns the chunked text.

For a first metered run, Jina: Read URL on any public page is the most informative: the response shows the extracted title, description, and markdown content, plus a usage figure reporting exactly what the call cost in tokens.

Node reference

11 nodes, generated from the connector itself so this reference always matches what ships.

Read and query nodes 5

Jina: Download Batch Output

jina_get_batch_output

Download the finished vectors of a completed batch embedding job.

Jina: Get Batch Job

jina_get_batch

Poll a batch embedding job's status and progress until it completes or fails.

Jina: Read URL

jina_read_url

Fetch any URL through Jina Reader (r.jina.ai) and return its title, description and main content as LLM-ready markdown, with token usage. Steering options control return format (markdown, html, text, screenshot, pageshot), CSS target selectors, token budget and image handling. Screenshots return as hosted URLs.

Jina: Search Web

jina_search_web

Search the web through Jina Search (s.jina.ai) and return results as structured JSON: title, URL, description and content per result, with site restriction and result count options.

Jina: Segment Text

jina_segment_text

Split long text into chunks sized for embedding models or LLM context windows. Free: costs zero tokens, yet still requires a valid API key, which makes it the connector's auth probe.

Action nodes 6

Jina: Cancel Batch Job

jina_cancel_batch

Cancel a running batch embedding job that is no longer needed.

Jina: Classify Text or Image

jina_classify

Classify text or image inputs against label names zero-shot, or against a trained classifier by id.

Jina: Create Batch Embedding Job

jina_create_batch_embeddings

Submit a large corpus as one asynchronous batch embedding job; returns a job id immediately. Poll with Jina: Get Batch Job and retrieve vectors with Jina: Download Batch Output.

Jina: Create Embeddings

jina_create_embeddings

Turn text or images into vectors synchronously. For thousands of inputs, use the batch embedding job family instead.

Jina: Rerank Documents

jina_rerank

Order candidate documents by relevance to a query, with scores. The quality step between a wide vector-search shortlist and the documents an LLM or person actually sees.

Jina: Train Classifier

jina_train_classifier

Train a classifier on labeled examples and return a classifier id that subsequent classify calls reference. Jina does not document listing or deleting classifiers, so store the returned id durably.

Using the integration

On the automation canvas

Jina nodes chain with every other node: Jina: Search Web can fan out into Jina: Read URL calls on the best hits, Jina: Rerank Documents can pick which knowledge article answers a question, and the results can land as comments on a Mattering Productivity task or records in a Mattering database.

From AI assistants over MCP

Every node speaks MCP, the open protocol for connecting AI assistants to tools. Claude, ChatGPT, and Matter can read pages, search the web, and rank documents through the same nodes your workflows use. Since nothing in the integration is destructive, the whole surface is safe to hand an agent; the only budget is tokens.

Steering Reader

Jina: Read URL exposes Jina's steering controls as plain config fields: return format (markdown, html, text, screenshot, pageshot), CSS target selectors, a token budget to cap how much of a large page is processed, and image handling. Screenshots come back as hosted URLs rather than raw bytes because the integration always requests Jina's JSON envelope, which also means responses are structured data, never a stream.

The retrieval pair

Jina: Create Embeddings and Jina: Rerank Documents cover the retrieval half of RAG: embed content into vectors once, then at question time rerank the candidate shortlist and pass only the top documents to a model. For chunking long documents ahead of embedding, Jina: Segment Text is free and makes chunk boundaries deliberate.

Classification, zero-shot or trained

Jina: Classify Text or Image labels inputs against label names with no training. When labels are stable and accuracy matters, Jina: Train Classifier fits a classifier on labeled examples and returns a classifier id that later classify calls reference. Jina does not document listing or deleting classifiers, so keep the returned id somewhere durable, such as a Mattering database record.

Batch embeddings as a create, poll, download chain

For thousands of texts, Jina: Create Batch Embedding Job submits one asynchronous job and returns its id. Jina: Get Batch Job polls status on a schedule, Jina: Download Batch Output retrieves the finished vectors, and Jina: Cancel Batch Job abandons a job early. For a handful of texts, synchronous Jina: Create Embeddings is simpler.

Troubleshooting

Test Connection fails with a 401
The key is invalid, was revoked, or picked up stray whitespace when pasted. Copy a fresh key from the jina.ai dashboard and update the credential. The connection test itself is free, so retesting costs nothing.
Metered nodes start failing while Segment Text still works
The token balance is exhausted. Jina: Segment Text is free so it keeps working, which makes it a useful tell: free node fine, metered nodes failing means balance, not auth. Top up the key's token balance at jina.ai.
Jina: Read URL returns little or wrong content for a page
Heavy client-side rendering, cookie walls, and unusual layouts can defeat default extraction. Use the target selector option to point Reader at the CSS selector holding the real content, or switch the return format to html to see what Reader actually received. For pages behind logins there is no fix: Reader fetches as an anonymous visitor.
A Reader call on a huge page consumed more tokens than expected
Reader meters by tokens processed, and large pages are large. Set the token budget option to cap processing per call, target a selector to skip boilerplate, and check the usage figure each response carries: it reports the exact spend, which a workflow can log or alert on.
Classify calls cannot find a trained classifier
Check that the classifier id is the exact value Jina: Train Classifier returned, on the same account as the key in use. Jina has no documented way to list classifiers, so a lost id cannot be rediscovered: store ids durably at training time, and retrain if one is gone.
A batch job seems stuck in progress
Large corpora legitimately take a while; keep the Jina: Get Batch Job poll on a relaxed schedule rather than hammering it. If a job is no longer needed, Jina: Cancel Batch Job stops it. Jobs that fail report a failure status on the poll rather than hanging forever, so a workflow branch on status covers both outcomes.
Calls fail with a 429 rate limit
Rate limits at Jina scale with the key's standing, and unauthenticated calls have the tightest limits of all, so first confirm the credential is attached and being used. Then space out scheduled sweeps and let workflows wait and retry. Search and Reader fan-outs are the usual culprits: a search that spawns twenty Reader calls at once is twenty times the request rate.

Jina AI is a trademark of its respective owner. Mattering uses platform names to identify the services these integrations connect to; no endorsement or affiliation is implied.

Start Free Explore our Products