Your agent's love query language

Query data across any API, database or file system with SQL. No custom integrations, no ETL, no glue code.

 brew install withcoral/tap/coral

SELECT i.title, p.url, p.author

FROM pagerduty.incidents i

JOIN github.pulls p

ON p.merged_at BETWEEN

i.started_at AND i.resolved_at

WHERE i.urgency = 'high';

title	url	author
checkout latency spike	acme/api/pull/1842	mchen
payment timeout cluster	acme/pay/pull/921	asingh
cart API 5xx burst	acme/cart/pull/447	dlee

— PagerDuty × GitHub · 3 rows · 84ms

— semantic hints applied · hot path cached

Turn every data source into a table. Query them together.

01Connect your sources

Point Coral at your APIs, databases and files. Each becomes a readonly schema.

coral source add github

coral source add linear

coral source add slack

— 3 sources · 18 tables

02Query across them with SQL

JOINs across any combination of sources. Coral handles auth, pagination, rate limits and schema mapping.

SELECT m.text, m.author, i.status

FROM slack.messages m

JOIN linear.issues i

ON m.text LIKE '%' || i.key || '%'

WHERE m.channel = '#engineering'

AND i.status != 'done'

ORDER BY m.ts DESC LIMIT 5;

— Slack × Linear · 3 rows · 190ms

03Plug it into any agent or framework

Use Coral over MCP or from the CLI. One runtime shared across agents.

codex mcp add coral -- coral mcp-stdio

claude mcp add coral -- coral mcp-stdio

npx skills add withcoral/skills

✓ One runtime shared across agents

04Turn usage into semantic context

Coral learns relationships, joins, and schema hints from every query.

coral sql "DESCRIBE EXTENDED

pagerduty.incidents"

columns

title Utf8 incident title

urgency Enum low | high

service Utf8 service key

relationships

service → datadog.service_health

service → github.deployments

started_at → github.pulls.merged_at

— 142 queries · 12 hints · 73% cache

coral source add github

coral source add linear

coral source add slack

coral sql "SELECT * FROM coral.tables LIMIT 6"

schema	table	description
github	pulls	Pull requests merged into a repo
github	issues	Issues and pull requests
linear	issues	Issues tracked in Linear
linear	projects	Projects and initiatives
slack	channels	Slack channels
slack	messages	Messages in a Slack channel

— 3 sources · 18 tables

SELECT m.text, m.author, i.status, i.priority

FROM slack.messages m

JOIN linear.issues i

ON m.text LIKE '%' || i.key || '%'

WHERE m.channel = '#engineering'

AND i.status != 'done'

ORDER BY m.ts DESC LIMIT 5;

— Slack × Linear · 3 rows · 190ms

codex mcp add coral -- coral mcp-stdio

claude mcp add coral -- coral mcp-stdio

npx skills add withcoral/skills

✓ Use Coral over MCP or from the CLI

✓ One runtime shared across agents

coral sql "DESCRIBE EXTENDED pagerduty.incidents"

pagerduty.incidents = "Active and historical incidents"

columns

title Utf8 incident title

urgency Enum low | high

service Utf8 canonical service key

started_at Timestamp incident start

resolved_at Timestamp incident end

relationships

service → datadog.service_health.service

service → github.deployments.service

started_at → github.pulls.merged_at (time window)

resolved_at → github.pulls.merged_at (time window)

recommended join

github.pulls.merged_at

BETWEEN pagerduty.incidents.started_at

AND pagerduty.incidents.resolved_at

— 142 queries · 12 schema hints · 73% cache hit rate

Connect your sources

Point Coral at your APIs, databases and files. Each becomes a readonly schema you can explore and query.

Query across them with SQL

Use standard SQL with JOINs across any combination of sources. Coral handles auth, pagination, rate limits and schema mapping.

Plug it into any agent or framework

Use Coral over MCP or from the CLI. One runtime shared across agents.

Turn usage into semantic context

Coral learns relationships, recommended joins, and schema hints from every query.

Works with the tools you already use

GitHub

GitLab

LaunchDarkly

Datadog

Sentry

OpenTelemetry

Slack

Intercom

Linear

ClickUp

Incident.io

Stripe

Agents are more accurate and cost efficient with Coral

We benchmarked Coral vs data source MCPs on the complex tasks that typify coding agent workloads. With Coral, Claude Code is ↑ 31% more accurate and ↓ 70% lower cost.

See the full methodology and results.

Directly With Coral

Accuracy +31%

Cost -70%

Built for production agent workloads

Read-only by design

Coral is a read layer, so agents can query across sources without mutating upstream systems. Safety without brittle sandboxing.

Permissions that fit agent workflows

Scoped tokens, workspace isolation, and per-source permissions. Give agents exactly the access they need and nothing more.

Efficient on every query

Query pushdown, caching, and efficient pagination keep queries responsive while reducing unnecessary API calls and token-heavy tool loops.

Adapts to real query patterns

Coral uses query patterns, schema access history, and source statistics to make discovery, caching, and execution better over time.

Give agents the context that changes outcomes

Unblock broken changes by combining PRs, CI failures, and linked issues in one query.

SELECT pr.number, pr.title, ci.failed_step, li.key AS linked_issue

FROM github.pulls pr

JOIN buildkite.builds ci ON ci.commit_sha = pr.head_sha

JOIN linear.issues li ON li.branch_name = pr.head_ref

WHERE ci.state = 'failed'

ORDER BY ci.finished_at DESC LIMIT 10;

SELECT sh.service, gh.sha, pd.title, sh.error_rate, s.mrr_at_risk

FROM datadog.service_health sh

JOIN github.deployments gh ON gh.service = sh.service

JOIN pagerduty.incidents pd ON pd.service = sh.service

JOIN stripe.service_revenue s ON s.service = sh.service

WHERE pd.status = 'triggered'

ORDER BY s.mrr_at_risk DESC, sh.error_rate DESC

LIMIT 10;

SELECT o.actor_email, o.event_type, a.resource, g.repo

FROM okta.system_logs o

JOIN aws.cloudtrail a ON a.actor_email = o.actor_email

JOIN github.audit_logs g ON g.actor_email = o.actor_email

WHERE o.risk_level = 'high'

ORDER BY o.timestamp DESC LIMIT 20;

actor_email	event_type	resource	repo
j.smith@acme.com	policy.lifecycle	arn:aws:s3:::backups	infra-terraform
m.chen@acme.com	user.session.start	arn:aws:iam::admin	deploy-scripts
j.smith@acme.com	app.oauth.consent	arn:aws:rds:::prod	infra-terraform

SELECT c.name, c.mrr, t.priority, s.sentiment, m.last_message_at

FROM stripe.customers c

JOIN zendesk.tickets t ON t.customer_id = c.id

JOIN support_inbox.sentiment s ON s.ticket_id = t.id

JOIN slack.customer_threads m ON m.customer_email = c.email

WHERE t.status = 'open'

ORDER BY c.mrr DESC LIMIT 10;

SELECT n.title, n.assignee, n.due_date, l.status, s.last_message_at

FROM notion.databases n

JOIN linear.issues l ON l.title = n.title

JOIN slack.messages s ON s.channel = n.slack_channel

WHERE n.status = 'in_progress'

AND n.due_date < CURRENT_DATE

ORDER BY n.due_date ASC LIMIT 10;

title	assignee	due_date	status	last_message_at
SSO setup for new hires	Dana Park	2025-03-08	in_progress	3 days ago
Vendor NDA review	Mike Chen	2025-03-06	todo	5 days ago
Q1 budget reforecast	Ava Singh	2025-03-04	in_progress	1 week ago

Get started in 60 seconds

Coral is open source under Apache 2.0. Install it, connect a source, and run your first query.

 brew install withcoral/tap/coral

Explore the docs Community GitHub

Developer FAQ

Why shouldn't I write API wrappers?

Wrappers solve one source at a time. Coral centralizes auth, retries, pagination, rate limiting, schema mapping, caching, and cross-source joins once for all agents.

How does Coral relate to MCP?

MCP provides a standard for connecting AI tools to data sources. Coral sits between your agents and your data sources, adding governance, cross-source joins, and caching on top. You can use Coral alongside MCP or as a standalone data layer — it provides a SQL interface that works with any agent framework.

Does Coral store my data?

Coral stores a small amount of contextual metadata locally and queries your APIs on demand. Hot paths are cached locally with a TTL.

Do I need to move data into a warehouse first?

No. Coral connects directly to your existing SaaS APIs, databases, and object stores. There's no ETL pipeline, no data migration, and no warehouse required. Your data stays where it is.

Is Coral an agent framework?

No. Coral is a data layer, not an agent framework. It provides a single SQL connection that any agent — regardless of framework — can use to access governed data across your systems.

Does my data leave my cloud?

No. Coral runs locally on your machine or infrastructure. Your data never leaves your environment — queries are executed against your sources directly and results stay on your infrastructure.

See more in Docs

What's next for Coral

In development

Self-host Coral for shared team use
Share source definitions, credentials, history, and schemas across a workspace
Explore schemas, query traces, and source health in a local UI
Data source actions as table functions

Planned next

Approval gated writes for agent actions
Coral as a managed cloud service
Enterprise SSO, audit logs, and org level permissions
Policy driven masking and redaction

number	title	failed_step	linked_issue
1842	reduce retry backoff	integration-tests	ENG-1291
1838	migrate checkout to v2 API	type-check	ENG-1287
1835	add rate limit to webhook	e2e-smoke	ENG-1280

service	deploy	incident	errors	mrr_at_risk
checkout-api	a3f29c1	checkout latency spike	12.4%	$182,000
payments-svc	e91d4b7	payment timeout cluster	8.7%	$141,000
cart-api	7c2f083	cart 5xx burst	6.1%	$96,000

name	mrr	priority	sentiment	last_message_at
Acme Corp	$12,400	urgent	negative	2 hours ago
Northwind Ltd	$8,200	high	neutral	6 hours ago
Contoso Inc	$6,800	urgent	negative	12 hours ago

text	author	status	priority
ENG-1291 is blocking the release	mchen	in_progress	urgent
anyone looking at ENG-1287?	dlee	todo	high
ENG-1280 fix is up for review	asingh	in_review	medium