Omni Analytics
Ask your company's governed Omni model a question in natural language and get warehouse data back. A three-step narrowing funnel — pick a model, pick a topic, run the query — with user permissions, access grants and row-level access filters documented as applying to every MCP query.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Omni Analytics via MCP
https://callbacks.omniapp.co/callback/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Omni Analytics Tools & Capabilities (3)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- The directory tool list is five short. Omni documents eight tools; Anthropic's 2026-08-16 snapshot names three. Read the vendor's tool reference, not the listing.
- The listed data tool is off by default.
getDatarequires Single shot query generation, which Omni states is disabled by default. A default organisation does not serve it. - The tool list is not a capability list. Disabled tools still appear in the client and fail at call time with
403 Feature is not enabled. - The funnel is effectively stateful.
pickTopicreadspickModel's choice andgetDataruns against both. Omni documents no behaviour for a wrong selection. - "Read" understates the documented surface.
askOmnican save model edits to a branch and create scheduled routines that deliver by email or Slack. - No tool annotations are published. Omni provides no annotation table and the endpoint is gated, so nothing here is vendor-labelled read-only.
- Permission enforcement is documented, not observed by us. Omni states MCP queries use the authenticated user's permissions; the gated endpoint meant we could not test it.
- API key connections use the key creator's permissions, not the caller's, unless impersonation headers are set — and
listBranchesignores those headers. - One header widens the boundary.
X-MCP-Query-All-Viewslets the AI query any view in the model rather than only curated topics. - It spends money twice. AI credits per interaction, and warehouse compute per query; Omni's
compute_routingnames MCP traffic under anagenticjob type. - Results are row-capped. 1,000 rows by default for new queries, configurable to 50,000;
runQuerydefaults to 500 and caps at 10,000. - Your data lands in the client. Omni states plainly that using the MCP Server puts your data in the tool you use it from, naming Cursor as its example.
- Multi-organization sign-in is fragile. The OAuth flow follows your most recent login cookie, so Omni instructs you to log out and back in first.
- OAuth tokens are invisible in the app. MCP OAuth PATs are not currently shown in Omni and must be read through the API.
- Admin gating is multi-layered. Enable AI, Omni Agent, the MCP server setting, per-tool capabilities, and Personal Access Tokens all have to line up.
- Pricing is not public. Omni publishes no pricing page, so plan-level entitlements could not be established.
Frequently asked questions
Does the Omni MCP server respect row-level security and user permissions?
Yes. Omni documents that every MCP query runs with the authenticated user's permissions, and that interactions are scoped to the model and topic supplied to the server. The modelling layer's controls apply underneath: access grants gate topics, views and individual fields, while access filters restrict rows by user attribute. An OAuth connection inherits your in-app permissions exactly.
Is the Omni MCP server stateful across pickModel, pickTopic and getData?
Yes, in effect. Omni documents `pickTopic` as returning topics in the model selected by `pickModel`, and `getData` as executing against the selected model and topic. The selection carries forward rather than travelling as arguments you can inspect. You can remove the ambiguity by pinning a model and topic with connection headers, which skips the picking steps entirely.
How many tools does the Omni Analytics MCP server have?
Anthropic's directory snapshot lists three. Omni's own tool reference documents eight: the three listed plus `runQuery`, `askOmni`, `checkStatus`, `listBranches` and `searchOmniDocs`. Which ones actually appear depends on organisation settings — `getData` and `runQuery` need Single shot query generation, which Omni states is disabled by default, so the listed set is not the served set.
Does using the Omni MCP server cost money?
Yes, twice over. Omni states that MCP Server usage consumes AI credits alongside its other AI features, with cost rising for complex multi-step analysis and larger data models. Separately, each query executes against your own warehouse. Omni's `compute_routing` setting names MCP traffic explicitly under an `agentic` job type so admins can route it to a dedicated warehouse.
What are models and topics in Omni's MCP tools?
A model is Omni's semantic layer over a database connection — tables, joins, fields and business logic in a three-layer structure. A topic is a curated dataset inside a model: a base table plus its joins, measures and dimensions, shaped for one business use case. Omni describes topics as shielding end-users from raw schemas. Together they bound what the agent can query.
Do I need an admin to enable the Omni MCP server?
Usually yes. Omni requires an Organization Admin to enable Enable AI, Omni Agent, and the MCP server setting, plus each individual tool capability. The MCP server setting is on by default; Single shot query generation, which powers `getData`, is not. Disabling MCP returns a 403 to new authorizations and blocks existing grants from calling tools.
Why is the Omni MCP endpoint on callbacks.omniapp.co rather than my own Omni URL?
Because it is a shared OAuth entry point, not your instance. Omni states this URL routes requests to your Omni organization after you identify yourself during the OAuth flow, which is why one host works for every customer. API key connections skip it and point at your own instance path instead. Omni warns the flow follows your most recent login cookie.
Sources
- Omni AI MCP Server overview — https://docs.omni.co/ai/mcp (fetched 2026-08-19 as Markdown via the
.mdsuffix) · retrieved 2026-08-19 - MCP server tools — https://docs.omni.co/ai/mcp/tools (fetched 2026-08-19) · retrieved 2026-08-19
- MCP authentication — https://docs.omni.co/ai/mcp/authentication (fetched 2026-08-19) · retrieved 2026-08-19
- AI MCP Server settings — https://docs.omni.co/ai/settings/mcp (fetched 2026-08-19) · retrieved 2026-08-19
- Using the MCP Server in Claude Desktop — https://docs.omni.co/ai/mcp/claude-desktop (fetched 2026-08-19) · retrieved 2026-08-19
- Using the MCP Server in Claude Code — https://docs.omni.co/ai/mcp/claude-code (fetched 2026-08-19) · retrieved 2026-08-19
- AI data security — https://docs.omni.co/ai/security (fetched 2026-08-19) · retrieved 2026-08-19
- Controlling data access using the modeling layer — https://docs.omni.co/modeling/develop/data-access-control (fetched 2026-08-19) · retrieved 2026-08-19
- Managing data access with connection permissions — https://docs.omni.co/administration/users/permissions (fetched 2026-08-19) · retrieved 2026-08-19
- Modeling in Omni — https://docs.omni.co/modeling (fetched 2026-08-19) · retrieved 2026-08-19
- Curating datasets with topics — https://docs.omni.co/modeling/topics (fetched 2026-08-19) · retrieved 2026-08-19
- Managing AI credit usage — https://docs.omni.co/ai/settings/usage (fetched 2026-08-19) · retrieved 2026-08-19
compute_routingmodel parameter — https://docs.omni.co/modeling/models/compute-routing (fetched 2026-08-19) · retrieved 2026-08-19default_row_limitmodel parameter — https://docs.omni.co/modeling/models/default-row-limit (fetched 2026-08-19) · retrieved 2026-08-19- Improving query performance with caching — https://docs.omni.co/analyze-explore/performance/caching (fetched 2026-08-19) · retrieved 2026-08-19
- Omni docs machine index — https://docs.omni.co/llms.txt (fetched 2026-08-19).
docs.omni.co/robots.txtpublishesContent-Signal: ai-train=yes, search=yes, ai-input=yes;omni.co/robots.txtallows all crawlers and publishes no Content-Signal axes · retrieved 2026-08-19 - Live anonymous probe — 2026-08-19: an
initializePOST tohttps://callbacks.omniapp.co/callback/mcpreturns HTTP 401 withwww-authenticate: Bearer resource_metadata="https://callbacks.omniapp.co/.well-known/oauth-protected-resource", scope="mcp:access", error="invalid_token"· retrieved 2026-08-19 - Live RFC 9728 check — all three protected-resource path forms probed 2026-08-19: root and path-append return identical 147-byte JSON descriptors naming
mcp:access; path-insert returns a 237 KB HTML application shell with HTTP 200 · retrieved 2026-08-19 - Live RFC 8414 authorization-server metadata — https://callbacks.omniapp.co/.well-known/oauth-authorization-server (fetched 2026-08-19; root and path-append forms identical,
S256only,scopes_supported: ["mcp:access"]) · retrieved 2026-08-19 - Anthropic connector directory — https://claude.ai/directory/8e7324aa-2671-46bf-9447-5bba8ccf8c2f (snapshot 2026-08-16) · retrieved 2026-08-16
- Omni pricing — https://omni.co/pricing returns HTTP 404 as of 2026-08-19; no public plan or entitlement information was available · retrieved 2026-08-19
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Data & Research
- Developer
- Omni Analytics
- Tools
- 3
- Domain
- callbacks.omniapp.co
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.