feat: two-layer rate limiting for MCP servers

server/cmd/gram/start.go

@@ -25,6 +25,7 @@ import (
 
 	"github.com/speakeasy-api/gram/server/internal/productfeatures"
 	"github.com/speakeasy-api/gram/server/internal/rag"
+	"github.com/speakeasy-api/gram/server/internal/ratelimit"
 
 	"github.com/speakeasy-api/gram/server/internal/about"
 	"github.com/speakeasy-api/gram/server/internal/agentworkflows"
@@ -652,10 +653,14 @@ func newStartCommand() *cli.Command {
 				return fmt.Errorf("failed to create mcp registry client: %w", err)
 			}
 
+			rateLimiter := ratelimit.New(redisClient, logger)
+			rateLimitConfigLoader := ratelimit.NewConfigLoader(logger, db, cache.NewRedisCacheAdapter(redisClient))
+
 			mux := goahttp.NewMuxer()
 			mux.Use(func(h http.Handler) http.Handler {
 				return otelhttp.NewHandler(h, "http", otelhttp.WithServerName("gram"))
 			})
+			mux.Use(middleware.RateLimitMiddleware(rateLimiter, rateLimitConfigLoader, logger))
 			mux.Use(middleware.CORSMiddleware(c.String("environment"), c.String("server-url"), chatSessionsManager))
 			mux.Use(middleware.NewHTTPLoggingMiddleware(logger))
 			mux.Use(customdomains.Middleware(logger, db, c.String("environment"), serverURL))
@@ -700,7 +705,7 @@ func newStartCommand() *cli.Command {
 			mcpMetadataService := mcpmetadata.NewService(logger, db, sessionManager, serverURL, siteURL, cache.NewRedisCacheAdapter(redisClient))
 			mcpmetadata.Attach(mux, mcpMetadataService)
 			externalmcp.Attach(mux, externalmcp.NewService(logger, tracerProvider, db, sessionManager, mcpRegistryClient))
-			mcp.Attach(mux, mcp.NewService(logger, tracerProvider, meterProvider, db, sessionManager, chatSessionsManager, env, posthogClient, serverURL, encryptionClient, cache.NewRedisCacheAdapter(redisClient), guardianPolicy, functionsOrchestrator, oauthService, billingTracker, billingRepo, telemSvc, productFeatures, ragService, temporalEnv), mcpMetadataService)
+			mcp.Attach(mux, mcp.NewService(logger, tracerProvider, meterProvider, db, sessionManager, chatSessionsManager, env, posthogClient, serverURL, encryptionClient, cache.NewRedisCacheAdapter(redisClient), guardianPolicy, functionsOrchestrator, oauthService, billingTracker, billingRepo, telemSvc, productFeatures, ragService, temporalEnv, rateLimiter), mcpMetadataService)
 			chat.Attach(mux, chat.NewService(logger, db, sessionManager, chatSessionsManager, openRouter, chatClient, posthogClient, telemSvc, assetStorage))
 			variations.Attach(mux, variations.NewService(logger, db, sessionManager))
 			customdomains.Attach(mux, customdomains.NewService(logger, db, sessionManager, &background.CustomDomainRegistrationClient{TemporalEnv: temporalEnv}))

server/database/schema.sql

@@ -634,6 +634,7 @@ CREATE TABLE IF NOT EXISTS toolsets (
   ),
   mcp_is_public BOOLEAN NOT NULL DEFAULT FALSE,
   mcp_enabled BOOLEAN NOT NULL DEFAULT FALSE,
+  rate_limit_rpm integer,
   tool_selection_mode TEXT NOT NULL DEFAULT 'static',
   custom_domain_id uuid,
 
@@ -1477,3 +1478,16 @@ WHERE deleted IS FALSE AND status = 'pending';
 -- Index for looking up invites by token (unconditional to prevent token reuse)
 CREATE UNIQUE INDEX IF NOT EXISTS team_invites_token_key
 ON team_invites (token);
+
+-- Platform-level rate limit overrides for Gram operators
+CREATE TABLE IF NOT EXISTS platform_rate_limits (
+  id uuid PRIMARY KEY DEFAULT gen_random_uuid(),
+  attribute_type text NOT NULL,
+  attribute_value text NOT NULL,
+  requests_per_minute integer NOT NULL,
+  created_at timestamptz NOT NULL DEFAULT clock_timestamp(),
+  updated_at timestamptz NOT NULL DEFAULT clock_timestamp(),
+  CONSTRAINT platform_rate_limits_attribute_type_value_key UNIQUE (attribute_type, attribute_value),
+  CONSTRAINT platform_rate_limits_attribute_type_check CHECK (attribute_type IN ('mcp_slug', 'project', 'organization')),
+  CONSTRAINT platform_rate_limits_requests_per_minute_check CHECK (requests_per_minute > 0)
+);

server/database/sqlc.yaml

@@ -385,3 +385,13 @@ sql:
         out: "../internal/externalmcp/repo"
         sql_package: "pgx/v5"
         omit_unused_structs: true
+
+  - schema: schema.sql
+    queries: ../internal/ratelimit/queries.sql
+    engine: postgresql
+    gen:
+      go:
+        package: "repo"
+        out: "../internal/ratelimit/repo"
+        sql_package: "pgx/v5"
+        omit_unused_structs: true

server/design/shared/toolset.go

@@ -84,6 +84,10 @@ var Toolset = Type("Toolset", func() {
 	Attribute("mcp_enabled", Boolean, "Whether the toolset is enabled for MCP")
 	Attribute("tool_selection_mode", String, "The mode to use for tool selection")
 	Attribute("custom_domain_id", String, "The ID of the custom domain to use for the toolset")
+	Attribute("rate_limit_rpm", Int, "Maximum requests per minute for this MCP server. When set, requests exceeding this limit receive a rate limit error.", func() {
+		Minimum(1)
+		Maximum(10000)
+	})
 	Attribute("external_oauth_server", ExternalOAuthServer, "The external OAuth server details")
 	Attribute("oauth_proxy_server", OAuthProxyServer, "The OAuth proxy server details")
 	Attribute("created_at", String, func() {
@@ -122,6 +126,10 @@ var ToolsetEntry = Type("ToolsetEntry", func() {
 	Attribute("mcp_enabled", Boolean, "Whether the toolset is enabled for MCP")
 	Attribute("tool_selection_mode", String, "The mode to use for tool selection")
 	Attribute("custom_domain_id", String, "The ID of the custom domain to use for the toolset")
+	Attribute("rate_limit_rpm", Int, "Maximum requests per minute for this MCP server. When set, requests exceeding this limit receive a rate limit error.", func() {
+		Minimum(1)
+		Maximum(10000)
+	})
 	Attribute("created_at", String, func() {
 		Description("When the toolset was created.")
 		Format(FormatDateTime)

server/design/toolsets/design.go

@@ -301,6 +301,10 @@ var UpdateToolsetForm = Type("UpdateToolsetForm", func() {
 	Attribute("mcp_is_public", Boolean, "Whether the toolset is public in MCP")
 	Attribute("custom_domain_id", String, "The ID of the custom domain to use for the toolset")
 	Attribute("tool_selection_mode", String, "The mode to use for tool selection")
+	Attribute("rate_limit_rpm", Int, "Maximum requests per minute for this MCP server. When set, requests exceeding this limit receive a rate limit error.", func() {
+		Minimum(1)
+		Maximum(10000)
+	})
 	security.ProjectPayload()
 	Required("slug")
 })