diff --git a/src/experiments.ts b/src/experiments.ts
index a7929494f99..1084b5324bb 100644
--- a/src/experiments.ts
+++ b/src/experiments.ts
@@ -147,6 +147,11 @@ export const ALL_EXPERIMENTS = experiments({
     default: true,
     public: false,
   },
+  mcpalpha: {
+    shortDescription: "Opt-in to early MCP features before they're widely released.",
+    default: false,
+    public: true,
+  },
   apptesting: {
     shortDescription: "Adds experimental App Testing feature",
     public: true,
diff --git a/src/mcp/prompts/dataconnect/index.ts b/src/mcp/prompts/dataconnect/index.ts
new file mode 100644
index 00000000000..e1c323ec73b
--- /dev/null
+++ b/src/mcp/prompts/dataconnect/index.ts
@@ -0,0 +1,9 @@
+import { isEnabled } from "../../../experiments";
+import { schema } from "./schema";
+import type { ServerPrompt } from "../../prompt";
+
+export const dataconnectPrompts: ServerPrompt[] = [];
+
+if (isEnabled("mcpalpha")) {
+  dataconnectPrompts.push(schema);
+}
diff --git a/src/mcp/prompts/dataconnect/schema.ts b/src/mcp/prompts/dataconnect/schema.ts
new file mode 100644
index 00000000000..cbc1dfd9028
--- /dev/null
+++ b/src/mcp/prompts/dataconnect/schema.ts
@@ -0,0 +1,73 @@
+import { prompt } from "../../prompt";
+import { loadAll } from "../../../dataconnect/load";
+import type { ServiceInfo } from "../../../dataconnect/types";
+import { BUILTIN_SDL, MAIN_INSTRUCTIONS } from "../../util/dataconnect/content";
+import { compileErrors } from "../../util/dataconnect/compile";
+
+function renderServices(fdcServices: ServiceInfo[]) {
+  if (!fdcServices.length) return "Data Connect Status: <UNCONFIGURED>";
+
+  return `\n\n## Data Connect Schema
+
+The following is the up-to-date content of existing schema files (their paths are relative to the Data Connect source directory).
+
+${fdcServices[0].schema.source.files?.map((f) => `\`\`\`graphql ${f.path}\n${f.content}\n\`\`\``).join("\n\n")}`;
+}
+
+function renderErrors(errors?: string) {
+  return `\n\n## Current Schema Build Errors\n\n${errors || "<NO ERRORS>"}`;
+}
+
+export const schema = prompt(
+  {
+    name: "schema",
+    description: "Generate or update your Firebase Data Connect schema.",
+    arguments: [
+      {
+        name: "prompt",
+        description:
+          "describe the schema you want generated or the edits you want to make to your existing schema",
+        required: true,
+      },
+    ],
+    annotations: {
+      title: "Generate Data Connect Schema",
+    },
+  },
+  async ({ prompt }, { config, projectId, accountEmail }) => {
+    const fdcServices = await loadAll(projectId, config);
+    const buildErrors = fdcServices.length
+      ? await compileErrors(fdcServices[0].sourceDirectory)
+      : "";
+
+    return [
+      {
+        role: "user" as const,
+        content: {
+          type: "text",
+          text: `
+${MAIN_INSTRUCTIONS}\n\n${BUILTIN_SDL}
+
+==== CURRENT ENVIRONMENT INFO ====
+
+User Email: ${accountEmail || "<NONE>"}
+Project ID: ${projectId || "<NONE>"}
+${renderServices(fdcServices)}${renderErrors(buildErrors)}
+
+==== USER PROMPT ====
+
+${prompt}
+
+==== TASK INSTRUCTIONS ====
+
+1. If Data Connect is marked as \`<UNCONFIGURED>\`, first run the \`firebase_init\` tool with \`{dataconnect: {}}\` arguments to initialize it.
+2. If there is not an existing schema to work with (or the existing schema is the commented-out default schema about a movie app), follow the user's prompt to generate a robust schema meeting the specified requirements.
+3. If there is already a schema, perform edits to the existing schema file(s) based on the user's instructions. If schema build errors are present and seem relevant to your changes, attempt to fix them.
+4. After you have performed edits on the schema, run the \`dataconnect_compile\` tool to build the schema and see if there are any errors. Fix errors that are related to the user's prompt or your changes.
+5. If there are errors, attempt to fix them. If you have attempted to fix them 3 times without success, ask the user for help.
+6. If there are no errors, write a brief paragraph summarizing your changes.`,
+        },
+      },
+    ];
+  },
+);
diff --git a/src/mcp/prompts/index.ts b/src/mcp/prompts/index.ts
index 78b4374e69d..9c15cd127a2 100644
--- a/src/mcp/prompts/index.ts
+++ b/src/mcp/prompts/index.ts
@@ -1,13 +1,14 @@
 import { ServerFeature } from "../types";
 import { ServerPrompt } from "../prompt";
 import { corePrompts } from "./core";
+import { dataconnectPrompts } from "./dataconnect";
 import { crashlyticsPrompts } from "./crashlytics";
 
 const prompts: Record<ServerFeature, ServerPrompt[]> = {
   core: corePrompts,
   firestore: [],
   storage: [],
-  dataconnect: [],
+  dataconnect: dataconnectPrompts,
   auth: [],
   messaging: [],
   remoteconfig: [],
diff --git a/src/mcp/tools/dataconnect/compile.ts b/src/mcp/tools/dataconnect/compile.ts
new file mode 100644
index 00000000000..e5904385051
--- /dev/null
+++ b/src/mcp/tools/dataconnect/compile.ts
@@ -0,0 +1,47 @@
+import { z } from "zod";
+import { tool } from "../../tool";
+import { pickService } from "../../../dataconnect/load";
+import { compileErrors } from "../../util/dataconnect/compile";
+
+export const compile = tool(
+  {
+    name: "build",
+    description:
+      "Use this to compile Firebase Data Connect schema, operations, and/or connectors and check for build errors.",
+    inputSchema: z.object({
+      error_filter: z
+        .enum(["all", "schema", "operations"])
+        .describe("filter errors to a specific type only. defaults to `all` if omitted.")
+        .optional(),
+      service_id: z
+        .string()
+        .optional()
+        .describe(
+          "The Firebase Data Connect service ID to look for. If omitted, builds all services defined in `firebase.json`.",
+        ),
+    }),
+    annotations: {
+      title: "Compile Data Connect",
+      readOnlyHint: true,
+    },
+    _meta: {
+      requiresProject: false,
+      requiresAuth: false,
+    },
+  },
+  async ({ service_id, error_filter }, { projectId, config }) => {
+    const serviceInfo = await pickService(projectId, config, service_id || undefined);
+    const errors = await compileErrors(serviceInfo.sourceDirectory, error_filter);
+    if (errors)
+      return {
+        content: [
+          {
+            type: "text",
+            text: `The following errors were encountered while compiling Data Connect from directory \`${serviceInfo.sourceDirectory}\`:\n\n${errors}`,
+          },
+        ],
+        isError: true,
+      };
+    return { content: [{ type: "text", text: "Compiled successfully." }] };
+  },
+);
diff --git a/src/mcp/tools/dataconnect/execute_graphql.ts b/src/mcp/tools/dataconnect/execute_graphql.ts
index d1d1c1ff6c2..2b70fa3ec46 100644
--- a/src/mcp/tools/dataconnect/execute_graphql.ts
+++ b/src/mcp/tools/dataconnect/execute_graphql.ts
@@ -3,9 +3,9 @@ import { z } from "zod";
 import { tool } from "../../tool";
 import * as dataplane from "../../../dataconnect/dataplaneClient";
 import { pickService } from "../../../dataconnect/load";
-import { graphqlResponseToToolResponse, parseVariables } from "./converter";
+import { graphqlResponseToToolResponse, parseVariables } from "../../util/dataconnect/converter";
 import { Client } from "../../../apiv2";
-import { getDataConnectEmulatorClient } from "./emulator";
+import { getDataConnectEmulatorClient } from "../../util/dataconnect/emulator";
 
 export const execute_graphql = tool(
   {
diff --git a/src/mcp/tools/dataconnect/execute_graphql_read.ts b/src/mcp/tools/dataconnect/execute_graphql_read.ts
index c73f896bc70..7f28d8a2daa 100644
--- a/src/mcp/tools/dataconnect/execute_graphql_read.ts
+++ b/src/mcp/tools/dataconnect/execute_graphql_read.ts
@@ -3,9 +3,9 @@ import { z } from "zod";
 import { tool } from "../../tool";
 import * as dataplane from "../../../dataconnect/dataplaneClient";
 import { pickService } from "../../../dataconnect/load";
-import { graphqlResponseToToolResponse, parseVariables } from "./converter";
+import { graphqlResponseToToolResponse, parseVariables } from "../../util/dataconnect/converter";
 import { Client } from "../../../apiv2";
-import { getDataConnectEmulatorClient } from "./emulator";
+import { getDataConnectEmulatorClient } from "../../util/dataconnect/emulator";
 
 export const execute_graphql_read = tool(
   {
diff --git a/src/mcp/tools/dataconnect/execute_mutation.ts b/src/mcp/tools/dataconnect/execute_mutation.ts
index ee4784489fc..59784f7c676 100644
--- a/src/mcp/tools/dataconnect/execute_mutation.ts
+++ b/src/mcp/tools/dataconnect/execute_mutation.ts
@@ -4,9 +4,9 @@ import { tool } from "../../tool";
 import { mcpError } from "../../util";
 import * as dataplane from "../../../dataconnect/dataplaneClient";
 import { pickService } from "../../../dataconnect/load";
-import { graphqlResponseToToolResponse, parseVariables } from "./converter";
+import { graphqlResponseToToolResponse, parseVariables } from "../../util/dataconnect/converter";
 import { Client } from "../../../apiv2";
-import { getDataConnectEmulatorClient } from "./emulator";
+import { getDataConnectEmulatorClient } from "../../util/dataconnect/emulator";
 
 export const execute_mutation = tool(
   {
diff --git a/src/mcp/tools/dataconnect/execute_query.ts b/src/mcp/tools/dataconnect/execute_query.ts
index f6248e82c41..4c58891f575 100644
--- a/src/mcp/tools/dataconnect/execute_query.ts
+++ b/src/mcp/tools/dataconnect/execute_query.ts
@@ -4,9 +4,9 @@ import { tool } from "../../tool";
 import { mcpError } from "../../util";
 import * as dataplane from "../../../dataconnect/dataplaneClient";
 import { pickService } from "../../../dataconnect/load";
-import { graphqlResponseToToolResponse, parseVariables } from "./converter";
+import { graphqlResponseToToolResponse, parseVariables } from "../../util/dataconnect/converter";
 import { Client } from "../../../apiv2";
-import { getDataConnectEmulatorClient } from "./emulator";
+import { getDataConnectEmulatorClient } from "../../util/dataconnect/emulator";
 
 export const execute_query = tool(
   {
diff --git a/src/mcp/tools/dataconnect/get_connector.ts b/src/mcp/tools/dataconnect/get_connector.ts
index 57db4958eb1..a3fe7eae1e5 100644
--- a/src/mcp/tools/dataconnect/get_connector.ts
+++ b/src/mcp/tools/dataconnect/get_connector.ts
@@ -3,7 +3,7 @@ import { tool } from "../../tool";
 import { toContent } from "../../util";
 import * as client from "../../../dataconnect/client";
 import { pickService } from "../../../dataconnect/load";
-import { connectorToText } from "./converter";
+import { connectorToText } from "../../util/dataconnect/converter";
 
 export const get_connectors = tool(
   {
diff --git a/src/mcp/tools/dataconnect/get_schema.ts b/src/mcp/tools/dataconnect/get_schema.ts
index bd8a8b66d87..05668c91708 100644
--- a/src/mcp/tools/dataconnect/get_schema.ts
+++ b/src/mcp/tools/dataconnect/get_schema.ts
@@ -3,7 +3,7 @@ import { tool } from "../../tool";
 import { toContent } from "../../util";
 import * as client from "../../../dataconnect/client";
 import { pickService } from "../../../dataconnect/load";
-import { schemaToText } from "./converter";
+import { schemaToText } from "../../util/dataconnect/converter";
 
 export const get_schema = tool(
   {
diff --git a/src/mcp/tools/dataconnect/index.ts b/src/mcp/tools/dataconnect/index.ts
index ee4e5755d5f..8e460728082 100644
--- a/src/mcp/tools/dataconnect/index.ts
+++ b/src/mcp/tools/dataconnect/index.ts
@@ -8,8 +8,10 @@ import { execute_graphql } from "./execute_graphql";
 import { execute_graphql_read } from "./execute_graphql_read";
 import { execute_query } from "./execute_query";
 import { execute_mutation } from "./execute_mutation";
+import { compile } from "./compile";
 
 export const dataconnectTools: ServerTool[] = [
+  compile,
   list_services,
   generate_schema,
   generate_operation,
diff --git a/src/mcp/util/dataconnect/compile.ts b/src/mcp/util/dataconnect/compile.ts
new file mode 100644
index 00000000000..27aee94c2e4
--- /dev/null
+++ b/src/mcp/util/dataconnect/compile.ts
@@ -0,0 +1,20 @@
+import { prettify } from "../../../dataconnect/graphqlError";
+import { DataConnectEmulator } from "../../../emulator/dataconnectEmulator";
+
+export async function compileErrors(
+  configDir: string,
+  errorFilter?: "all" | "schema" | "operations",
+) {
+  const errors = (await DataConnectEmulator.build({ configDir })).errors;
+  return (
+    errors
+      ?.filter((e) => {
+        const isOperationError = ["query", "mutation"].includes(e.path?.[0] as string);
+        if (errorFilter === "operations") return isOperationError;
+        if (errorFilter === "schema") return !isOperationError;
+        return true;
+      })
+      .map(prettify)
+      .join("\n") || ""
+  );
+}
diff --git a/src/mcp/util/dataconnect/content.ts b/src/mcp/util/dataconnect/content.ts
new file mode 100644
index 00000000000..417d72a73a0
--- /dev/null
+++ b/src/mcp/util/dataconnect/content.ts
@@ -0,0 +1,653 @@
+export const MAIN_INSTRUCTIONS = `
+Closely follow the following instructions:
+
+You are Firebase Data Connect expert that is responsible for creating data connect schemas code in GraphQL for users.You will be given a description of the desired schema using Firebase Data Connect and your task is to write the schema code in GraphQL that fulfills the requirements and correct any mistakes in your generation.
+
+For example, if I were to ask for a schema for a GraphQL database that contains a table called "users" with a field called "name" and another table called "posts" with a field called "body", I would get the following schema:
+\`\`\`
+type User @table {
+  name: String!
+}
+
+type Post @table {
+  body: String!
+  author: User
+}
+\`\`\`
+
+Simple Firebase Data Connect schema often takes the following form:
+\`\`\`graphql
+type TableName @table {
+  uuidField: UUID
+  uuidArrayField: [UUID]
+  stringField: String
+  stringArrayField: [String]
+  intField: Int
+  intArrayField: [Int]
+  int64Field: Int64
+  int64ArrayField: [Int64]
+  floatField: Float
+  floatArrayField: [Float]
+  booleanField: Boolean
+  booleanArrayField: [Boolean]
+  timestampField: Timestamp
+  timestampArrayField: [Timestamp]
+  dateField: Date
+  dateArrayField: [Date]
+  vectorField: Vector @col(size:168)
+}
+\`\`\`
+
+Leave out objects named after \`Query\` and \`Mutation\`
+
+Firebase Data Connect implicitly adds \`id: UUID!\` to every table and implicitly makes it primary key. Therefore, leave out the \`id\` field.
+
+Use \`UUID\` type instead of \`ID\` type or \`String\` type for id-like fields.
+
+Array reference fields, like \`[SomeTable]\` and \`[SomeTable!]!\`, are not supported. Use the singular reference field instead.
+For example, for a one-to-many relationship like one user is assiend to many bugs in a software project:
+\`\`\`graphql
+type User @table {
+  name: String!
+  # bugs: [Bug]   # Not supported. Do not use
+}
+
+type Bug @table {
+  title: String!
+  assignee: User
+  reporter: User
+}
+\`\`\`
+
+For another example, for a many-to-many relationship like each crew member is assigned to many chores and each chores requires many crews to complete:
+\`\`\`graphql
+type Crew @table {
+  name: String!
+  # assignedChores: [Chore!]!  # No supported. Do not use
+}
+
+type Chore @table {
+  name: String!
+  description: String!
+  # assignedCrews: [Crews!]!  # No supported. Do not use
+}
+
+type Assignment @table(key: ["crew", "chore"]) {
+  crew: Crew!
+  chore: Chore!
+}
+\`\`\`
+
+Leave out \`@relation\` because it is not supported yet.
+
+Leave out \`directive\`, \`enum\` and \`scalar\`.
+
+Leave out \`@view\`.
+
+Be sure that your response contains a valid Firebase Data Connect schema in a single GraphQL code block inside of triple backticks and closely follows my instructions and description.
+`.trim();
+
+export const BUILTIN_SDL = `
+# Directives
+
+Directives define specific behaviors that can be applied to fields or types within a GraphQL schema.
+
+## Data Connect Defined
+
+### @col on \`FIELD_DEFINITION\` {:#col}
+Customizes a field that represents a SQL database table column.
+
+Data Connect maps scalar Fields on [\`@table\`](directive.md#table) type to a SQL column of
+corresponding data type.
+
+- scalar [\`UUID\`](scalar.md#UUID) maps to [\`uuid\`](https://www.postgresql.org/docs/current/datatype-uuid.html).
+- scalar [\`String\`](scalar.md#String) maps to [\`text\`](https://www.postgresql.org/docs/current/datatype-character.html).
+- scalar [\`Int\`](scalar.md#Int) maps to [\`int\`](https://www.postgresql.org/docs/current/datatype-numeric.html).
+- scalar [\`Int64\`](scalar.md#Int64) maps to [\`bigint\`](https://www.postgresql.org/docs/current/datatype-numeric.html).
+- scalar [\`Float\`](scalar.md#Float) maps to [\`double precision\`](https://www.postgresql.org/docs/current/datatype-numeric.html).
+- scalar [\`Boolean\`](scalar.md#Boolean) maps to [\`boolean\`](https://www.postgresql.org/docs/current/datatype-boolean.html).
+- scalar [\`Date\`](scalar.md#Date) maps to [\`date\`](https://www.postgresql.org/docs/current/datatype-datetime.html).
+- scalar [\`Timestamp\`](scalar.md#Timestamp) maps to [\`timestamptz\`](https://www.postgresql.org/docs/current/datatype-datetime.html).
+- scalar [\`Any\`](scalar.md#Any) maps to [\`jsonb\`](https://www.postgresql.org/docs/current/datatype-json.html).
+- scalar [\`Vector\`](scalar.md#Vector) maps to [\`pgvector\`](https://github.com/pgvector/pgvector).
+
+Array scalar fields are mapped to [Postgres arrays](https://www.postgresql.org/docs/current/arrays.html).
+
+###### Example: Serial Primary Key
+
+For example, you can define auto-increment primary key.
+
+\`\`\`graphql
+type Post @table {
+  id: Int! @col(name: "post_id", dataType: "serial")
+}
+\`\`\`
+
+Data Connect converts it to the following SQL table schema.
+
+\`\`\`sql
+CREATE TABLE "public"."post" (
+  "post_id" serial NOT NULL,
+  PRIMARY KEY ("id")
+)
+\`\`\`
+
+###### Example: Vector
+
+\`\`\`graphql
+type Post @table {
+  content: String! @col(name: "post_content")
+  contentEmbedding: Vector! @col(size:768)
+}
+\`\`\`
+
+| Argument | Type | Description |
+|---|---|---|
+| \`name\` | [\`String\`](scalar.md#String) | The SQL database column name. Defaults to snake_case of the field name. |
+| \`dataType\` | [\`String\`](scalar.md#String) | Configures the custom SQL data type.  Each GraphQL type can map to multiple SQL data types. Refer to [Postgres supported data types](https://www.postgresql.org/docs/current/datatype.html).  Incompatible SQL data type will lead to undefined behavior. |
+| \`size\` | [\`Int\`](scalar.md#Int) | Required on [\`Vector\`](scalar.md#Vector) columns. It specifies the length of the Vector. \`textembedding-gecko@003\` model generates [\`Vector\`](scalar.md#Vector) of \`@col(size:768)\`. |
+
+### @default on \`FIELD_DEFINITION\` {:#default}
+Specifies the default value for a column field.
+
+For example:
+
+\`\`\`graphql
+type User @table(key: "uid") {
+  uid: String! @default(expr: "auth.uid")
+  number: Int! @col(dataType: "serial")
+  createdAt: Date! @default(expr: "request.time")
+  role: String! @default(value: "Member")
+  credit: Int! @default(value: 100)
+}
+\`\`\`
+
+The supported arguments vary based on the field type.
+
+| Argument | Type | Description |
+|---|---|---|
+| \`value\` | [\`Any\`](scalar.md#Any) | A constant value validated against the field's GraphQL type during compilation. |
+| \`expr\` | [\`Any_Expr\`](scalar.md#Any_Expr) | A CEL expression whose return value must match the field's data type. |
+| \`sql\` | [\`Any_SQL\`](scalar.md#Any_SQL) | A raw SQL expression, whose SQL data type must match the underlying column.  The value is any variable-free expression (in particular, cross-references to other columns in the current table are not allowed). Subqueries are not allowed either. See [PostgreSQL defaults](https://www.postgresql.org/docs/current/sql-createtable.html#SQL-CREATETABLE-PARMS-DEFAULT) for more details. |
+
+### @index on \`FIELD_DEFINITION\` | \`OBJECT\` {:#index}
+Defines a database index to optimize query performance.
+
+\`\`\`graphql
+type User @table @index(fields: ["name", "phoneNumber"], order: [ASC, DESC]) {
+    name: String @index
+    phoneNumber: Int64 @index
+    tags: [String] @index # GIN Index
+}
+\`\`\`
+
+##### Single Field Index
+
+You can put [\`@index\`](directive.md#index) on a [\`@col\`](directive.md#col) field to create a SQL index.
+
+\`@index(order)\` matters little for single field indexes, as they can be scanned
+in both directions.
+
+##### Composite Index
+
+You can put \`@index(fields: [...])\` on [\`@table\`](directive.md#table) type to define composite indexes.
+
+\`@index(order: [...])\` can customize the index order to satisfy particular
+filter and order requirement.
+
+| Argument | Type | Description |
+|---|---|---|
+| \`name\` | [\`String\`](scalar.md#String) | Configure the SQL database index id.  If not overridden, Data Connect generates the index name: - \`{table_name}_{first_field}_{second_field}_aa_idx\` - \`{table_name}_{field_name}_idx\` |
+| \`fields\` | [\`[String!]\`](scalar.md#String) | Only allowed and required when used on a [\`@table\`](directive.md#table) type. Specifies the fields to create the index on. |
+| \`order\` | [\`[IndexFieldOrder!]\`](enum.md#IndexFieldOrder) | Only allowed for \`BTREE\` [\`@index\`](directive.md#index) on [\`@table\`](directive.md#table) type. Specifies the order for each indexed column. Defaults to all \`ASC\`. |
+| \`type\` | [\`IndexType\`](enum.md#IndexType) | Customize the index type.  For most index, it defaults to \`BTREE\`. For array fields, only allowed [\`IndexType\`](enum.md#IndexType) is \`GIN\`. For [\`Vector\`](scalar.md#Vector) fields, defaults to \`HNSW\`, may configure to \`IVFFLAT\`. |
+| \`vector_method\` | [\`VectorSimilarityMethod\`](enum.md#VectorSimilarityMethod) | Only allowed when used on vector field. Defines the vector similarity method. Defaults to \`INNER_PRODUCT\`. |
+
+### @ref on \`FIELD_DEFINITION\` {:#ref}
+Defines a foreign key reference to another table.
+
+For example, we can define a many-to-one relation.
+
+\`\`\`graphql
+type ManyTable @table {
+  refField: OneTable!
+}
+type OneTable @table {
+  someField: String!
+}
+\`\`\`
+Data Connect adds implicit foreign key column and relation query field. So the
+above schema is equivalent to the following schema.
+
+\`\`\`graphql
+type ManyTable @table {
+  id: UUID! @default(expr: "uuidV4()")
+  refField: OneTable! @ref(fields: "refFieldId", references: "id")
+  refFieldId: UUID!
+}
+type OneTable @table {
+  id: UUID! @default(expr: "uuidV4()")
+  someField: UUID!
+  # Generated Fields:
+  # manyTables_on_refField: [ManyTable!]!
+}
+\`\`\`
+Data Connect generates the necessary foreign key constraint.
+
+\`\`\`sql
+CREATE TABLE "public"."many_table" (
+  "id" uuid NOT NULL DEFAULT uuid_generate_v4(),
+  "ref_field_id" uuid NOT NULL,
+  PRIMARY KEY ("id"),
+  CONSTRAINT "many_table_ref_field_id_fkey" FOREIGN KEY ("ref_field_id") REFERENCES "public"."one_table" ("id") ON DELETE CASCADE
+)
+\`\`\`
+
+###### Example: Traverse the Reference Field
+
+\`\`\`graphql
+query ($id: UUID!) {
+  manyTable(id: $id) {
+    refField { id }
+  }
+}
+\`\`\`
+
+###### Example: Reverse Traverse the Reference field
+
+\`\`\`graphql
+query ($id: UUID!) {
+  oneTable(id: $id) {
+    manyTables_on_refField { id }
+  }
+}
+\`\`\`
+
+##### Optional Many-to-One Relation
+
+An optional foreign key reference will be set to null if the referenced row is deleted.
+
+In this example, if a \`User\` is deleted, the \`assignee\` and \`reporter\`
+references will be set to null.
+
+\`\`\`graphql
+type Bug @table {
+  title: String!
+  assignee: User
+  reproter: User
+}
+
+type User @table { name: String!  }
+\`\`\`
+
+##### Required Many-to-One Relation
+
+A required foreign key reference will cascade delete if the referenced row is
+deleted.
+
+In this example, if a \`Post\` is deleted, associated comments will also be
+deleted.
+
+\`\`\`graphql
+type Comment @table {
+  post: Post!
+  content: String!
+}
+
+type Post @table { title: String!  }
+\`\`\`
+
+##### Many To Many Relation
+
+You can define a many-to-many relation with a join table.
+
+\`\`\`graphql
+type Membership @table(key: ["group", "user"]) {
+  group: Group!
+  user: User!
+  role: String! @default(value: "member")
+}
+
+type Group @table { name: String! }
+type User @table { name: String! }
+\`\`\`
+
+When Data Connect sees a table with two reference field as its primary key, it
+knows this is a join table, so expands the many-to-many query field.
+
+\`\`\`graphql
+type Group @table {
+  name: String!
+  # Generated Fields:
+  # users_via_Membership: [User!]!
+  # memberships_on_group: [Membership!]!
+}
+type User @table {
+  name: String!
+  # Generated Fields:
+  # groups_via_Membership: [Group!]!
+  # memberships_on_user: [Membership!]!
+}
+\`\`\`
+
+###### Example: Traverse the Many-To-Many Relation
+
+\`\`\`graphql
+query ($id: UUID!) {
+  group(id: $id) {
+    users: users_via_Membership {
+      name
+    }
+  }
+}
+\`\`\`
+
+###### Example: Traverse to the Join Table
+
+\`\`\`graphql
+query ($id: UUID!) {
+  group(id: $id) {
+    memberships: memberships_on_group {
+      user { name }
+      role
+    }
+  }
+}
+\`\`\`
+
+##### One To One Relation
+
+You can even define a one-to-one relation with the help of [\`@unique\`](directive.md#unique) or \`@table(key)\`.
+
+\`\`\`graphql
+type User @table {
+  name: String
+}
+type Account @table {
+  user: User! @unique
+}
+# Alternatively, use primary key constraint.
+# type Account @table(key: "user") {
+#   user: User!
+# }
+\`\`\`
+
+###### Example: Transerse the Reference Field
+
+\`\`\`graphql
+query ($id: UUID!) {
+  account(id: $id) {
+    user { id }
+  }
+}
+\`\`\`
+
+###### Example: Reverse Traverse the Reference field
+
+\`\`\`graphql
+query ($id: UUID!) {
+  user(id: $id) {
+    account_on_user { id }
+  }
+}
+\`\`\`
+
+##### Customizations
+
+- \`@ref(constraintName)\` can customize the SQL foreign key constraint name (\`table_name_ref_field_fkey\` above).
+- \`@ref(fields)\` can customize the foreign key field names.
+- \`@ref(references)\` can customize the constraint to reference other columns.
+   By default, \`@ref(references)\` is the primary key of the [\`@ref\`](directive.md#ref) table.
+   Other fields with [\`@unique\`](directive.md#unique) may also be referred in the foreign key constraint.
+
+| Argument | Type | Description |
+|---|---|---|
+| \`constraintName\` | [\`String\`](scalar.md#String) | The SQL database foreign key constraint name. Defaults to snake_case \`{table_name}_{field_name}_fkey\`. |
+| \`fields\` | [\`[String!]\`](scalar.md#String) | Foreign key fields. Defaults to \`{tableName}{PrimaryIdName}\`. |
+| \`references\` | [\`[String!]\`](scalar.md#String) | The fields that the foreign key references in the other table. Defaults to its primary key. |
+
+### @table on \`OBJECT\` {:#table}
+Defines a relational database table.
+
+In this example, we defined one table with a field named \`myField\`.
+
+\`\`\`graphql
+type TableName @table {
+  myField: String
+}
+\`\`\`
+Data Connect adds an implicit \`id\` primary key column. So the above schema is equivalent to:
+
+\`\`\`graphql
+type TableName @table(key: "id") {
+  id: String @default(expr: "uuidV4()")
+  myField: String
+}
+\`\`\`
+
+Data Connect generates the following SQL table and CRUD operations to use it.
+
+\`\`\`sql
+CREATE TABLE "public"."table_name" (
+  "id" uuid NOT NULL DEFAULT uuid_generate_v4(),
+  "my_field" text NULL,
+  PRIMARY KEY ("id")
+)
+\`\`\`
+
+ * You can lookup a row: \`query ($id: UUID!) { tableName(id: $id) { myField } } \`
+ * You can find rows using: \`query tableNames(limit: 20) { myField }\`
+ * You can insert a row: \`mutation { tableName_insert(data: {myField: "foo"}) }\`
+ * You can update a row: \`mutation ($id: UUID!) { tableName_update(id: $id, data: {myField: "bar"}) }\`
+ * You can delete a row: \`mutation ($id: UUID!) { tableName_delete(id: $id) }\`
+
+##### Customizations
+
+- \`@table(singular)\` and \`@table(plural)\` can customize the singular and plural name.
+- \`@table(name)\` can customize the Postgres table name.
+- \`@table(key)\` can customize the primary key field name and type.
+
+For example, the \`User\` table often has a \`uid\` as its primary key.
+
+\`\`\`graphql
+type User @table(key: "uid") {
+  uid: String!
+  name: String
+}
+\`\`\`
+
+ * You can securely lookup a row: \`query { user(key: {uid_expr: "auth.uid"}) { name } } \`
+ * You can securely insert a row: \`mutation { user_insert(data: {uid_expr: "auth.uid" name: "Fred"}) }\`
+ * You can securely update a row: \`mutation { user_update(key: {uid_expr: "auth.uid"}, data: {name: "New Name"}) }\`
+ * You can securely delete a row: \`mutation { user_delete(key: {uid_expr: "auth.uid"}) }\`
+
+[\`@table\`](directive.md#table) type can be configured further with:
+
+ - Custom SQL data types for columns. See [\`@col\`](directive.md#col).
+ - Add SQL indexes. See [\`@index\`](directive.md#index).
+ - Add SQL unique constraints. See [\`@unique\`](directive.md#unique).
+ - Add foreign key constraints to define relations. See [\`@ref\`](directive.md#ref).
+
+| Argument | Type | Description |
+|---|---|---|
+| \`name\` | [\`String\`](scalar.md#String) | Configures the SQL database table name. Defaults to snake_case like \`table_name\`. |
+| \`singular\` | [\`String\`](scalar.md#String) | Configures the singular name. Defaults to the camelCase like \`tableName\`. |
+| \`plural\` | [\`String\`](scalar.md#String) | Configures the plural name. Defaults to infer based on English plural pattern like \`tableNames\`. |
+| \`key\` | [\`[String!]\`](scalar.md#String) | Defines the primary key of the table. Defaults to a single field named \`id\`. If not present already, Data Connect adds an implicit field \`id: UUID! @default(expr: "uuidV4()")\`. |
+
+### @unique on \`FIELD_DEFINITION\` | \`OBJECT\` {:#unique}
+Defines unique constraints on [\`@table\`](directive.md#table).
+
+For example,
+
+\`\`\`graphql
+type User @table {
+    phoneNumber: Int64 @unique
+}
+type UserProfile @table {
+    user: User! @unique
+    address: String @unique
+}
+\`\`\`
+
+- [\`@unique\`](directive.md#unique) on a [\`@col\`](directive.md#col) field adds a single-column unique constraint.
+- [\`@unique\`](directive.md#unique) on a [\`@table\`](directive.md#table) type adds a composite unique constraint.
+- [\`@unique\`](directive.md#unique) on a [\`@ref\`](directive.md#ref) defines a one-to-one relation. It adds unique constraint
+   on \`@ref(fields)\`.
+
+[\`@unique\`](directive.md#unique) ensures those fields can uniquely identify a row, so other [\`@table\`](directive.md#table)
+type may define \`@ref(references)\` to refer to fields that has a unique constraint.
+
+| Argument | Type | Description |
+|---|---|---|
+| \`indexName\` | [\`String\`](scalar.md#String) | Configures the SQL database unique constraint name.  If not overridden, Data Connect generates the unique constraint name: - \`table_name_first_field_second_field_uidx\` - \`table_name_only_field_name_uidx\` |
+| \`fields\` | [\`[String!]\`](scalar.md#String) | Only allowed and required when used on OBJECT, this specifies the fields to create a unique constraint on. |
+
+### @view on \`OBJECT\` {:#view}
+Defines a relational database Raw SQLview.
+
+Data Connect generates GraphQL queries with WHERE and ORDER BY clauses.
+However, not all SQL features has native GraphQL equivalent.
+
+You can write **an arbitrary SQL SELECT statement**. Data Connect
+would map Graphql fields on [\`@view\`](directive.md#view) type to columns in your SELECT statement.
+
+* Scalar GQL fields (camelCase) should match a SQL column (snake_case)
+  in the SQL SELECT statement.
+* Reference GQL field can point to another [\`@table\`](directive.md#table) type. Similar to foreign key
+  defined with [\`@ref\`](directive.md#ref) on a [\`@table\`](directive.md#table) type, a [\`@view\`](directive.md#view) type establishes a relation
+  when \`@ref(fields)\` match \`@ref(references)\` on the target table.
+
+In this example, you can use \`@view(sql)\` to define an aggregation view on existing
+table.
+
+\`\`\`graphql
+type User @table {
+  name: String
+  score: Int
+}
+type UserAggregation @view(sql: """
+  SELECT
+    COUNT(*) as count,
+    SUM(score) as sum,
+    AVG(score) as average,
+    PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY score) AS median,
+    (SELECT id FROM "user" LIMIT 1) as example_id
+  FROM "user"
+""") {
+  count: Int
+  sum: Int
+  average: Float
+  median: Float
+  example: User
+  exampleId: UUID
+}
+\`\`\`
+
+###### Example: Query Raw SQL View
+
+\`\`\`graphql
+query {
+  userAggregations {
+    count sum average median
+    exampleId example { id }
+  }
+}
+\`\`\`
+
+##### One-to-One View
+
+An one-to-one companion [\`@view\`](directive.md#view) can be handy if you want to argument a [\`@table\`](directive.md#table)
+with additional implied content.
+
+\`\`\`graphql
+type Restaurant @table {
+  name: String!
+}
+type Review @table {
+  restaurant: Restaurant!
+  rating: Int!
+}
+type RestaurantStats @view(sql: """
+  SELECT
+    restaurant_id,
+    COUNT(*) AS review_count,
+    AVG(rating) AS average_rating
+  FROM review
+  GROUP BY restaurant_id
+""") {
+  restaurant: Restaurant @unique
+  reviewCount: Int
+  averageRating: Float
+}
+\`\`\`
+
+In this example, [\`@unique\`](directive.md#unique) convey the assumption that each \`Restaurant\` should
+have only one \`RestaurantStats\` object.
+
+###### Example: Query One-to-One View
+
+\`\`\`graphql
+query ListRestaurants {
+  restaurants {
+    name
+    stats: restaurantStats_on_restaurant {
+      reviewCount
+      averageRating
+    }
+  }
+}
+\`\`\`
+
+###### Example: Filter based on One-to-One View
+
+\`\`\`graphql
+query BestRestaurants($minAvgRating: Float, $minReviewCount: Int) {
+  restaurants(where: {
+    restaurantStats_on_restaurant: {
+      averageRating: {ge: $minAvgRating}
+      reviewCount: {ge: $minReviewCount}
+    }
+  }) { name }
+}
+\`\`\`
+
+##### Customizations
+
+- One of \`@view(sql)\` or \`@view(name)\` should be defined.
+  \`@view(name)\` can refer to a persisted SQL view in the Postgres schema.
+- \`@view(singular)\` and \`@view(plural)\` can customize the singular and plural name.
+
+[\`@view\`](directive.md#view) type can be configured further:
+
+ - [\`@unique\`](directive.md#unique) lets you define one-to-one relation.
+ - [\`@col\`](directive.md#col) lets you customize SQL column mapping. For example, \`@col(name: "column_in_select")\`.
+
+##### Limitations
+
+Raw SQL view doesn't have a primary key, so it doesn't support lookup. Other
+[\`@table\`](directive.md#table) or [\`@view\`](directive.md#view) cannot have [\`@ref\`](directive.md#ref) to a view either.
+
+View cannot be mutated. You can perform CRUD operations on the underlying
+table to alter its content.
+
+**Important: Data Connect doesn't parse and validate SQL**
+
+- If the SQL view is invalid or undefined, related requests may fail.
+- If the SQL view return incompatible types. Firebase Data Connect may surface
+  errors.
+- If a field doesn't have a corresponding column in the SQL SELECT statement,
+  it will always be \`null\`.
+- There is no way to ensure VIEW to TABLE [\`@ref\`](directive.md#ref) constraint.
+- All fields must be nullable in case they aren't found in the SELECT statement
+  or in the referenced table.
+
+**Important: You should always test [\`@view\`](directive.md#view)!**
+
+| Argument | Type | Description |
+|---|---|---|
+| \`name\` | [\`String\`](scalar.md#String) | The SQL view name. If neither \`name\` nor \`sql\` are provided, defaults to the snake_case of the singular type name. \`name\` and \`sql\` cannot be specified at the same time. |
+| \`sql\` | [\`String\`](scalar.md#String) | SQL \`SELECT\` statement used as the basis for this type. SQL SELECT columns should use snake_case. GraphQL fields should use camelCase. \`name\` and \`sql\` cannot be specified at the same time. |
+| \`singular\` | [\`String\`](scalar.md#String) | Configures the singular name. Defaults to the camelCase like \`viewName\`. |
+| \`plural\` | [\`String\`](scalar.md#String) | Configures the plural name. Defaults to infer based on English plural pattern like \`viewNames\`. |
+`.trim();
diff --git a/src/mcp/tools/dataconnect/converter.ts b/src/mcp/util/dataconnect/converter.ts
similarity index 100%
rename from src/mcp/tools/dataconnect/converter.ts
rename to src/mcp/util/dataconnect/converter.ts
diff --git a/src/mcp/tools/dataconnect/emulator.ts b/src/mcp/util/dataconnect/emulator.ts
similarity index 100%
rename from src/mcp/tools/dataconnect/emulator.ts
rename to src/mcp/util/dataconnect/emulator.ts