Jedify Production

Query a governed semantic layer over Snowflake, BigQuery or Databricks in plain language, read-only SQL.

Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.

Paste it into any MCP client. Setup docs

What the Jedify Production connector does

The Jedify MCP server connects a governed semantic layer over your data warehouse to Claude, ChatGPT and any MCP-compatible agent. Jedify sits between an agent and Snowflake, BigQuery or Databricks, mapping business terms — revenue, churn, a metric definition — onto warehouse tables, so an agent asks in plain language rather than SQL. Sign-in is OAuth or an API key.

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.

Jedify Production tools (16)

  • ask_a_research_question
  • ask_a_single_question
  • ask_an_agent
  • ask_questions
  • check_question_status
  • contextualize
  • create_share_link
  • data_collection_strategy
  • get_questions_data
  • get_semantic_layer_entities
  • list_agents
  • logout
  • run_sql_query
  • search_semantic_layer
  • stop_question
  • wait_for_completion

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • The tool count depends on a URL parameter, not on the connector. Anthropic lists 16 Asker-mode tools; Jedify documents 25 across three modes, selected by ?mode= or a proxy flag. A tool census of the listed endpoint does not describe an installation that connected with a different string.
  • No application scopes exist. The RFC 9728 descriptor at https://be.jedify.com/.well-known/oauth-protected-resource/mcp declares scopes_supported as an
  • Anthropic's directory records the auth posture wrongly. The entry sets is_authless: true and auth_posture: "no_auth". The listed endpoint returned HTTP 401 with a WWW-Authenticate: Bearer challenge to an anonymous request on 2026-08-23. Jedify requires a login.
  • Three listed tools have no vendor documentation. ask_an_agent, list_agents and create_share_link appear in Anthropic's listing and in none of Jedify's three tool references. We cannot describe what ask_an_agent reaches, and it is dispatcher-shaped by name.
  • We could not read tool schemas or safety annotations. The listed endpoint returned 401, so we never saw a tools/list response and cannot report readOnlyHint or destructiveHint for any tool. The read/write split above is our classification of names against Jedify's descriptions.
  • /mcp/message initializes without authentication. A single anonymous initialize returned HTTP 200 with the server's full instruction block on 2026-08-23, while the listed /mcp path returned 401. Only the handshake is open — we did not attempt any tool call, and tool access remains credentialed on Jedify's documented account.
  • No tool can observe cost. No tool in any of the 25 names carries a credit, balance, usage, quota or billing verb. Warehouse query cost is real — Jedify's own Athena page notes per-TB-scanned billing — and nothing in the surface reports it.
  • Published apps flatten permissions. Jedify warns that a published dashboard uses one API key for every viewer, so all viewers see data at that key's level rather than their own.
  • Row-level security needs SSO. Jedify's Snowflake session-variable route for per-user row filtering "requires SSO." Without it, an admin must map each user to a Snowflake role by hand.
  • Jedify publishes no numeric rate limit. Its REST reference documents a 429 response but states no threshold, and documents no idempotency key on any question or function-creation call.
  • Jedify disclaims output accuracy in writing. Its disclaimer states AI-generated outputs "may occasionally contain errors, omissions, or misinterpretations," that validating them is the user's responsibility, and that Jedify "does not guarantee the correctness, completeness, or suitability of any result."
  • Whether a share link can be revoked is unresolved. We found no Jedify documentation on expiring or withdrawing a share link, and we did not create one.

Frequently asked questions

How many tools does the Jedify MCP server have?

Sixteen in Anthropic's listing, but 25 in Jedify's documentation. Jedify's MCP server runs three modes — Asker, Editor and Builder — and its docs state the mode changes which tools your client sees. The directory lists Asker only. Editor adds six semantic-layer editing tools and Builder adds six function-management tools, and a URL parameter selects the mode.

Can the Jedify connector change my semantic layer or my data?

Not in the default Asker mode, and only as a proposal in Editor mode. Jedify documents that Editor mode never changes your semantic layer directly — sl_commit_edits stages a changeset as review-pending sessions and a human approves before anything applies. Asker mode's only write is create_share_link. No mode writes to your warehouse.

Can the Jedify connector run arbitrary SQL against my warehouse?

Yes, read-only SELECT statements. The run_sql_query tool is in Anthropic's listing and Jedify's Asker reference describes it as a read-only SQL SELECT against the data source. The server's own instructions state that only SELECT is allowed, that INSERT, UPDATE, DELETE, DROP, ALTER and CREATE are barred, and that queries must be single-line.

What OAuth permissions does the Jedify MCP server request?

None that narrow anything. Jedify's RFC 9728 resource descriptor declares scopes_supported as an empty array, and its authorization server advertises only email, profile and phone — all OpenID identity claims. Checked live on 2026-08-23. So the consent screen offers no way to grant asking a question without also granting read-only SQL.

Does the Jedify connector need a login?

Yes, despite Anthropic's directory recording it as authless. The directory entry sets is_authless to true and auth_posture to no_auth, but an anonymous request to the listed endpoint returned HTTP 401 with a Bearer challenge on 2026-08-23. Jedify documents browser sign-in for desktop clients and an API key for hosted agents.

What data can the Jedify connector read?

Whatever your warehouse holds, filtered by your permissions. Jedify connects to Snowflake, BigQuery, Databricks, Redshift and AWS Athena, and its authentication page states access is scoped to the signed-in user's or key's permissions. For Snowflake, Jedify forwards the end-user identity so native row access policies filter rows per user.

Why does a published Jedify dashboard show everyone the same data?

Because it uses one API key for all viewers. Jedify warns that a published app uses one API key for everyone who opens it, so every viewer sees data at that key's permission level rather than their own. Jedify's guidance for sensitive data such as HR or finance is to pick a key whose access matches what all viewers should see.

Why are the Jedify Builder tools missing from my client?

Because you are in Asker mode, which is the default. Jedify's troubleshooting page states that Builder mode requires the --builder flag in the proxy args, and that an API key carries its own mode. When using a key, Jedify's documented fix is to set the server URL to https://be.jedify.com/mcp/message?mode=builder, which takes precedence over the key's mode.

Sources

Use it in an agent

Put Jedify Production to work.

Connect Jedify Production once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.

Call (650) 285-1019Our AI receptionist answers.