Does Curated Data replace our RAG stack or vector database?

Not necessarily. Retrieval finds relevant material; Curated Data adds the ownership, approval state, classification, and revision history that decide which high-consequence knowledge an agent should trust. You can use it as a governed lane alongside a broader RAG corpus.

Do I need to write the format by hand?

No. You write concept docs in a normal markdown editor with a frontmatter form. Curated Data validates the structure, resolves links, and handles the Open Knowledge Format details for you. You can also import existing markdown in bulk.

Who can approve changes?

Members with a reviewer or admin role. Every edit stays in draft until a reviewer approves it, and the approval is recorded in the audit log alongside who made the change and when — so nothing your agents read was skipped past your team.

What exactly do my agents receive?

Only each document’s currently approved revision. Drafts and in-review changes are never served, and an approved change becomes available immediately without a separate workspace publish step.

How is one company’s knowledge kept separate from another’s?

Every workspace is an isolated tenant. API tokens are scoped to a single company, so a token can only ever read that company’s approved knowledge.

Which agents can connect over MCP?

Any MCP-capable agent, client, or model runtime. You add one server entry with your workspace’s hosted same-origin MCP URL and a bearer token. Anything that isn’t MCP-native can use the REST v1 API instead.

When does an approved edit become live?

Immediately. Approval promotes that document’s revision to current, and connected REST and MCP clients see it on their next request. Other documents are unaffected, and the revision history remains available for audit.

Is there an API rate limit?

Yes — request limits scale by plan and are shared by every API token in an organization. Responses include RateLimit-Policy and RateLimit headers; exhausted temporary windows return Retry-After.

How Curated Data works

The short version of how curation flows from edit to approval, how your approved knowledge reaches your agents, and the open format it's built on.

I build the agent

Connect MCP or REST

Use a hosted endpoint, inspect the tools and response shape, and test a scoped token.

Builder path →

I own the knowledge

Run the review workflow

Import or write concepts, assign an owner, review the diff, and publish one approved revision.

Knowledge-owner path →

I review the controls

Trace the access boundary

See where workspace, token, classification, and approval status are enforced before retrieval.

Governance path →

Quickstart

Get started in 10 minutes

From an empty workspace to an agent answering from your approved knowledge: five steps, no infrastructure to run.

1
Create your workspace
Sign up free and name your organization. You get the full curate-review-approve workflow, MCP, and REST from the start.
2
Write your first concept
Add a concept doc for something your agents keep getting wrong — a metric definition, a policy, a table’s meaning — in a markdown editor with a frontmatter form. Or bulk-import existing markdown.
3
Review and publish it
Move the draft into review. A reviewer sees the diff and approves it — recorded in the audit log and immediately live to authorized agents. Small teams can self-approve; larger ones require a second pair of eyes.
4
Mint a scoped token
In Settings → API Tokens, create a ckd_live_ token scoped to the read access and classifications your agent should have.
5
Connect an agent
Add one MCP config block, or call the REST v1 API. Your agent now retrieves each document’s approved revision at answer time.

Start free

Trust

Why your agents can rely on it

Agents hallucinate when they cannot reach your knowledge; they fill the gap from training data and guesswork. Curated Data closes that gap: your agents retrieve knowledge your team has actually vetted, at answer time. A unit of knowledge is a concept doc: the fields in a structured header (owner, status, type, tags), the explanation in plain prose, and explicit links to related concepts. Because each one carries an owner and an approval status, an agent's answer is traceable back to a specific, reviewed source. It won't make an agent perfect, but it shrinks the room for guessing. Here's a rendered concept doc:

policies/refunds.md

Approved

Ownercs-leadership

Typepolicy

Tagsbillingcustomer-facingrefunds

Refunds

Self-serve refunds are available within 30 days of purchase on every plan, including annual. After 30 days, offer prorated account credit instead; credits need manager approval before they are issued.

Quick rules

0 to 30 daysfull refund to the original payment method
31+ daysprorated account credit, manager approval
chargebacksescalate to billing · never self-serve

Linked pagesaccount creditannual planscancellations

Workflow

Edit → review → approve → live

Anyone on your team can draft or edit a concept doc. Changes stay in draft until a reviewer looks at them. A reviewer approves and publishes the change, which records who signed off and when — so nothing reaches an agent without review.

Approval promotes only that document's revision to current. It is immediately available over REST and MCP; there is no separate whole-workspace publish action and unrelated documents do not change.

Delivery

Serving your agents over MCP and REST

Your currently approved documents are served two ways from the same source of truth. The hosted MCP server lets any MCP-capable agent connect with a single same-origin config entry and a per-company token — nothing to run or scale yourself. The REST v1 API is served from the same origin and covers anything that isn't MCP-native.

Tokens are scoped to one company, support rotation, and are rate-limited by plan. Agents only ever see approved knowledge — never a draft or revision still in review.

Format

Built on the Open Knowledge Format

Underneath, every concept doc is written in the Open Knowledge Format — an open standard for describing what a company knows. Each doc is a markdown file with a YAML frontmatter header; its path, like tables/orders.md, is its stable identity, and links written as [customers](customers.md) connect docs into a knowledge graph.

Because the format is open, your knowledge stays portable and yours — no lock-in. Curated Data adds the curation, approvals, hosting, and serving that turn it into something your agents can trust.

Ready to try it?

Create a workspace and connect an agent

Free to start — curate a concept, get it approved, then connect over MCP or REST.

Start free Connect over MCP →

Developer quickstart

Connect an agent two ways — a hosted MCP server or the REST v1 API. Both serve the same approved document revisions. Compatible MCP clients sign in with browser OAuth and receive a short-lived, organization-scoped token. REST clients and MCP clients without OAuth use a scoped API token created in Settings → Tokens. A token is restricted to its selected workspace, scopes, and classifications.

Connect over MCP

bash — Claude Code with OAuth

$ claude mcp add --transport http curated-data \
    https://curateddata.megacorp.company/api/curated-data/mcp

# Run /mcp, choose curated-data, then sign in and approve the organization.

bash — local stdio bridge via npx (uses a scoped API token)

$ claude mcp add --transport stdio \
    --env CURATED_DATA_API_TOKEN=ckd_live_•••••••• \
    curated-data -- npx -y curated-data-mcp

# Works in any MCP client that launches stdio servers (Claude Code, Cursor, VS Code).

Or call the REST API

bash — retrieve an approved concept

$ curl https://curateddata.megacorp.company/api/curated-data/v1/concepts/tables/orders \
    -H "Authorization: Bearer ckd_live_••••••••"

# → the approved, versioned concept + its resolved links

Endpoints

`GET /api/curated-data/v1/concepts`	List the workspace’s currently approved concepts.
`GET /api/curated-data/v1/concepts/{path}`	Fetch one concept with its resolved [[links]].
`GET /api/curated-data/v1/search?q=...`	Retrieve the concepts most relevant to a query, at answer time.
`POST /api/curated-data/v1/imports`	Start a bulk import job from a source you point it at (write-scoped).
`GET /api/curated-data/v1/imports/{jobId}`	Check an import job's status.
`GET /api/curated-data/v1/reviews`	List the review queue for an approve-scoped token; approve or reject a revision from the same surface.
`GET /api/curated-data/v1/bundle`	Download the org's latest published bundle, once one has been published.

REST requests and MCP clients without OAuth need an Authorization: Bearer header. Rate limits scale by plan; Enterprise limits and availability commitments are set in a sales agreement. Responses include usage headers.

Full reference: every route, request, and response shape is described in the OpenAPI contract at /api/curated-data/v1/openapi.json.

Production checklist

Before the first agent goes live

Start small enough to inspect the entire knowledge path. A first production domain is ready when the team can answer each item below.

The first agent’s high-consequence questions are listed.
Each required concept has a named owner.
The current revision was reviewed and approved.
Classifications match the agent’s intended audience.
The token has only the scopes the agent needs.
Drafts and restricted concepts fail retrieval tests.
Answers include a source path or citation your team can inspect.
Someone owns corrections after launch.

Walk through the demo →Explore example concepts →

Frequently asked questions

Still have a question?

Sales can help with rollout, limits, and Enterprise requirements.

Talk to sales

Get product updates

New endpoints, MCP tools, and format changes — no spam.