Share & serve agent-ready data

Supacharge Supabase for Analytics & Automation

DataMCP builds a context layer and serves it alongside query results via MCP.

Join the waitlist See how it works

Works with Claude, Cursor, and any MCP client.

agent session — via datamcp
agent> list customer segments that need re-activation
→ datamcp: resolving "customer" → entity: Customer (deduplicated across
crm.accounts, billing.customers, product.workspaces)
→ datamcp: resolving "re-activation" → metric: churn_risk_score · v3 · owner: growth
→ datamcp: applying contract: pii.email masked · freshness ≤ 24h ✓
segment customers last_active churn_risk
───────────────────────── ───────── ─────────── ──────────
Paid, no login 30d 1,204 2026-08-14 0.81
Trial expired, high usage 388 2026-08-29 0.74
Enterprise, seat drop >20% 41 2026-09-02 0.69
agent> draft a re-activation sequence for segment 1
→ datamcp: 1,204 rows · 3 columns · 412 tokens (raw schema would be 1,480)
1 MCP endpoint · 0 raw schema dumps

Outcomes

Point your agents at data that actually makes sense

DataMCP gives agents a semantic layer with defined concepts, governed metrics and clean entities.

automations

Fuel automations

Agents run workflows on data they actually understand and can trust. Remove the guesswork out of your data.

metrics

Unify your metrics

One governed definition per metric, expressed in Python, YAML, and SQL. Share metrics that finally agree with each other.

quality

Fix data issues at the source

Surface and fix broken, duplicated, and drifting data before an agent confidently hallucinates on top of it.

bi

Better BI, fewer tokens

Queries resolve against defined concepts instead of raw schema dumps. Cheaper, faster, and dramatically more accurate.

entities

Entity resolution built in

Deduplicate key entities like person and company across sources, so GTM and finance agents talk about the same thing.

access

MCP-native access

Everything served over the Model Context Protocol. Any MCP client, whether Claude, your own agents, or your IDE, plugs straight in.

How it works

From raw tables to agent-ready data

DataMCP turns your database into a semantic layer agents can trust.

  1. 01

    scan

    Connect your warehouse. DataMCP inventories every table, column, and pipeline.

  2. 02

    rank

    Signal from noise: usage, lineage, and freshness rank what actually matters.

  3. 03

    profile

    Distributions, nulls, drift. A statistical fingerprint of every dataset.

  4. 04

    define

    Concepts get standard definitions; metrics compile to Python, YAML & SQL.

  5. 05

    classify

    PII, sensitivity, and domain tags applied automatically and kept current.

  6. 06

    contract

    Definitions become enforced contracts. Break one and hear about it first.

information_schema.columns Before
table_name column_name data_type
accounts acct_email_addr text
accounts created_ts int8
customers_v2 email varchar
customers_v2 churn_score_old float4
workspaces owner_email text
tmp_customers_bak email_1 text
… 214 more rows, 3 duplicate email columns
customer.contract.yml After
entity: Customer
sources: [crm.accounts, billing.customers,
product.workspaces]
columns:
+ email: { pii: true, mask: always }
metrics:
churn_risk_score:
version: 3 · owner: growth
compiles_to: [sql, python, yaml]
checks:
+ freshness: { max_age: 24h }

Get the receipts

DataMCP measures the tokens you save and the analytical accuracy you gain.

−72%

tokens per agent query vs raw schema access

3.4×

faster answers on BI and analytics questions

98%

metric queries resolved to a governed definition

1

MCP endpoint for your whole data estate (coming soon)

Measured on internal benchmark workloads. Your numbers will differ; DataMCP reports them per query.

Give your agents data that makes sense.

DataMCP is in early access for Supabase teams.

See how it works