DataPulse MY — a practical introduction

What you will learn: what DataPulse MY is, the problem it solves, and a hands-on feel for its capabilities by driving the live MCP server through real queries.

By the end you should be able to answer three questions:

1. What is DataPulse MY?
2. What does it do that a raw open-data portal does not?
3. How do you ask it questions and how do you read its answers?

The services and dataset identifiers named below are stable. Their health state is not: DataPulse re-probes on a schedule, so freshness values, record counts, and statuses change continuously. This page deliberately does not pin those numbers; it shows the shape of each answer and points you to where you can read the live values yourself.

• Live catalogue and health: https://www.data-pulse.my/health/latest.json
• Live tool schemas and value semantics: the MCP reference
• Live aggregate freshness: call get_freshness_summary() (shown below)

---

1. What DataPulse MY is

DataPulse MY is an open, read-only trust and evidence layer for Malaysian public datasets. It continuously watches the official datasets published by data.gov.my, BNM, DOSM, DOE, KKM, KPDN, MET Malaysia and similar agencies, and answers one question those agencies' own portals do not:

Is this dataset fresh, and can I prove what state it was in when I looked at it?

A data portal tells you a CSV exists. DataPulse tells you — with machine-checked evidence — that the dataset is reachable, how old its newest row is relative to the cadence its publisher declares, whether its shape or record count has drifted, and it packages that observation so a third party can verify it independently rather than taking DataPulse's word.

One sentence: DataPulse MY turns "Malaysian open data exists" into "Malaysian open data can be trusted at a point in time."

The catalogue's dataset count is maintained as a live manifest (read the authoritative number from datapulse.json or the dataset_total field of health/latest.json). Each dataset is classified against a 10-status taxonomy (fresh | aging | stale | discontinued | degraded | browser_dependent | unreachable | unknown | unknown_freshness | reference).

---

2. Why this exists

Open-data portals publish files; they rarely tell you whether a file is current. A shared CSV can sit stale for months. An AI agent that answers from it will confidently quote outdated figures. DataPulse exists so that "the number says X" becomes "the number says X, and the source was last fresh on this date, with this evidence."

It is intentionally read-only. It does not write to any upstream agency. It observes, classifies, signs its observations, and serves them — so consumers get a verdict they can act on and a trail they can check.

For the full freshness methodology and the meaning of each status, see health-methodology.

---

3. Tour of the core capabilities

The MCP server exposes 18 read-only tools in four clusters. For each, this section shows the call and explains the shape of what comes back — every value is an illustration of structure, not a current claim. Read the MCP reference for exact schemas and read the live endpoints for current values.

3.1 Find datasets — search_datasets

Ask in natural language what you are looking for; DataPulse returns ranked matches, each with an id, title, source, licence, and a freshness status.

TOOL: search_datasets(query: "fuel price")

What you get back is a list shaped like this (ids, titles, sources and licenses are stable catalogue facts; the per-row status is live and changes):

The rows are ordered by relevance (a score field). For a fuel-price query you should expect fuelprice (Malaysian Fuel Prices) and pricecatcher (KPDN grocery prices) among the top matches, plus price-index datasets from DOSM — but which exact set is ranked first reflects the live catalogue at query time.

search_datasets also accepts optional source, licence, and limit filters, useful when you must restrict an answer to a specific agency or to content you may reuse under a specific licence.

3.2 Inspect one dataset — get_dataset

Drill into a single dataset's stored record.

TOOL: get_dataset(dataset_id: "fuelprice")

The response carries the dataset's identity, license, custodianship, and refresh contract (stable) alongside its current freshness verdict (live). For fuelprice you will reliably see fields like:

• name: Malaysian Fuel Prices
• steward / attribution: Ministry of Finance Malaysia via data.gov.my
• refresh_frequency: weekly
• licence: Open Government Licence (Malaysia)
• url: the catalogue URL on data.gov.my
• status: the live verdict (fresh/aging/… at runtime)
• staleness_days: days since the newest observed row (live)
• record_count: the observed row count, and whether it is inside the expected-tolerance band (live)
• anomaly_detected: whether DataPulse's freshness-anomaly detection flagged an irregular gap (live)

This is how you get the "is it current" question answered with attributable fields — the identity fields are stable, the freshness fields are the live observation.

3.3 Verify one dataset — verify_dataset

This is DataPulse's differentiator. verify_dataset returns the dataset record plus its health observation plus the Sigstore bundle references that let you prove the evidence was not tampered with.

TOOL: verify_dataset(dataset_id: "fuelprice")

The response is the same get_dataset payload re-grouped into dataset (identity), health (live observation), and evidence (the machine-checked fields), augmented with:

• signed + verifier_output: the live result of an in-process cosign verify-blob-attestation against the served Sigstore bundle. When it runs, signed is true and verifier_output quotes Verified OK. If the verifier cannot run, the tool fails closed — signed is false rather than silently claiming a validity it could not check.
• bundle_ref / statement_ref / provenance_artifact_url: the served HTTPS artifact URLs for this dataset's signed evidence.
• verification_hint: an exact cosign verify-blob-attestation command you can run yourself against those artifacts.

Because the bundle is served as a normal HTTPS artifact, you can verify DataPulse's evidence with the same tooling used for software supply-chain signatures — and the live MCP already does this on your behalf. You do not have to trust DataPulse; you can check it either way.

3.4 Summarise health coverage

Quick, aggregate questions about the whole catalogue:

• get_freshness_summary() returns the total dataset count and the count per freshness status as of the latest published snapshot — always current when you call it.
• find_stale(), find_anomalies(), find_deteriorating(), find_recovering(), find_unreliable(), find_schema_drift() return the datasets meeting each criterion, ranked with evidence.
• trust_verdict() and verify_attestation() expose the trust-plane posture.
• get_provenance() and get_evidence() return citation-ready provenance and evidence receipts.

Call get_freshness_summary() yourself — it is the single fastest way to see the current state of the whole catalogue in one line:

TOOL: get_freshness_summary()
# returns e.g.: fresh=N, aging=N, stale=N, reference=N; dataset_total=N
# checked_at=<the latest observation timestamp>

The exact counts change continuously; the field names and the checked_at timestamp are the parts you should rely on.

---

4. Worked pattern: "find something current about fuel"

Put the pieces together. To answer "what fuel-related datasets can I cite against, and is the main one fresh?"

1. search_datasets("fuel price") → narrows to a ranked set you can inspect. Expect a fuel-price catalogue entry at or near the top.
2. get_dataset(<that id>) → read its identity + live freshness verdict (status, staleness, cadence).
3. verify_dataset(<same id>) → get the evidence artifact URLs so the claim is independently checkable.

That is the whole shape of a defensible answer: find → inspect → verify.

---

5. Where to go next

• MCP reference — the generated schema and signature for every one of the 18 tools; the exact field semantics.
• Live health — the current freshness snapshot these tools read (and no-auth, so you can compare tool output to the raw JSON).
• Agent quickstart — five-minute onboarding for AI agents that must cite Malaysian data correctly.
• Agent workflows — deeper integration patterns.
• Evidence receipt spec — the format of the signed evidence artefacts.
• The trust-layer notebook — a Colab tutorial that verifies a receipt end-to-end with cosign.

6. Try it yourself

The server needs no auth key. In a client that speaks MCP, connect to https://mcp.data-pulse.my/mcp and run search_datasets on a topic you care about (try geography, health, economy, transport, or an agency you follow). Then get_dataset the top result and read its identity and freshness verdict. Then verify_dataset the same id and inspect the evidence artifact URLs and verification_hint it returns. Every value you see is the live current state — no number in this page will disagree with what the tools answer, because this page does not pin any.