SmartGateSmartGate

Features

Three pillars of the MCP-native intelligence layer — not another LLM proxy.

What SmartGate is

SmartGate is the intelligence layer between AI and the world: an MCP-native algorithmic gateway that helps humans and agents research, understand, remember, and act with confidence. On the product surface you get three pillars—Token Control, Traffic Customs, and Audit & Compliance—backed by five intelligence capabilities (Research, Context, Memory, Control, Pipeline) powered by sevensmart_* primitives on one Streamable HTTP MCP endpoint.

Your host model still reasons and chats. SmartGate sits on the tool path: search and fetch from the open web, deduplicate overlapping passages, compress context before it hits the model window, enforce hard budget caps, and write team-scoped audit rows for every invocation. That is gateway-side intelligence for agent traffic—not a second chatbot and not a model router.

What SmartGate is not

Per the Positioning Bible, SmartGate isnot another AI assistant, a ChatGPT alternative, an AI search engine, anLLM proxy, or a model router. We do not replace your model provider or terminate chat completions. We govern what agents fetch, search, compress, deduplicate, remember, and budget-cap before that traffic inflates the host-model bill.

Marketing proof on this site is gateway-local (for example one public URL cleaned to Markdown, or Usage Reports from audited tool rows). We do not invent customer logos or third-party ROI studies on these pages. When you need numbers, readPricingand the signed-in dashboard—not invented case studies.

How the pillars map to the runtime

Token Controlis the Control capability in product language:smart_context_gate for configurable compression, smart_dedup for overlapping chunks, andsmart_budget_guard for check / count / hard-cap. Use it when agent loops silently compound tokens and alerts alone are not enforcement.

Traffic Customsis Research + Pipeline on the wire:smart_search,smart_fetch, andsmart_pipe templates (research, read, remember) so builders stop re-implementing fetch/search glue per agent. Uncontrolled fetch is the upstream cause of token bloat; shape traffic once at the gateway.

Audit & Complianceis the observability floor for production agents: every MCP and REST invocation writes a team-scoped, trace-linked record you can query and export. That is governance of tool traffic—not a claim of healthcare or banking certification packs.

Memory (smart_memory) spans pillars: team-scoped facts that survive the next session, not a public encyclopedia. Pair it with Pipeline when you want research → extract → remember as one call instead of glue code. Full tool semantics live inDocs; role-shaped journeys live underSolutions.

  • Token Control

    Agent loops inflate context every iteration. Intercept at the tool layer with smart_context_gate, smart_dedup, and smart_budget_guard — then prove savings in Usage Reports.

    Learn more

  • Traffic Customs

    Uncontrolled fetch and search are the upstream cause of token bloat. Aggregate, extract, deduplicate, and compress — as one pipe or step-by-step.

    Learn more

  • Audit & Compliance

    Every MCP and REST invocation writes an audit record — team-scoped, trace-linked, exportable. Minimum bar for production agents in 2026.

    Learn more

Use cases

Two interfaces, one runtime: Web and Playground for people who should not see MCP JSON; MCP and API for builders who need research primitives, context engineering, memory, budget control, and pipeline orchestration without rebuilding infrastructure for every agent.

  • Agent builders

    Wire one Streamable HTTP MCP endpoint into Cursor, Claude Desktop, or OpenClaw — compress and cap tool output without rebuilding fetch/search plumbing per agent.

  • FinOps & platform teams

    Hard-cap token spend at the tool layer, then prove savings from audit rows — not from host-model invoices alone.

  • Security & compliance reviewers

    Answer who fetched what, when, and at what token cost — team-scoped activity logs with export for reviews.

  • Individuals & researchers

    Research and decisions without pasting four redundant searches into chat — see personal scenes under Solutions.

The seven primitives, and when not to use them

SmartGate exposes seven smart_* tools on one Streamable HTTP MCP endpoint (the same operations are reachable over 17 REST endpoints across 7 groups: audit, budget, compress, dedup, fetch, memory, search). The purpose and the when not to use line under each tool are copied from the tool metadata the gateway serves to MCP clients — not paraphrased for this page.

SmartGate MCP tools, purpose, when to use and when not to use
ToolPurposeUse it whenDo not use it when
smart_searchWeb search · read-onlyRun multi-engine web search and aggregate snippets.Research is needed and no definitive URL is available yet.The user already supplied the exact URL — use smart_fetch instead.
smart_fetchFetch URL as Markdown · read-onlyFetch a public HTTP(S) URL and return Markdown-friendly content.The user provides a link and needs the page body or article text.Full text is already in the conversation, or the target is not a fetchable URL.
smart_context_gateCompress context · read-onlyCompress long prompts/context (LLMLingua) before the host LLM call.Context is too large for the model budget and text is already available.Short messages or already small context. Optional purpose: Pass a short goal (step intent, user query) to pre-filter paragraphs by relevance before compression — e.g. "API rate limits", "pricing tables".
smart_dedupSemantic deduplication · read-onlySemantic deduplication across multiple text passages (SemHash).Several chunks or search results overlap heavily.A single string or unrelated documents.
smart_budget_guardBudget check/count/record · read-onlyTeam token budget check, count, or usage record.action=check or count before expensive tools; action=record after token-consuming work.General chat, reasoning, or as a substitute for the host model.
smart_memoryTeam memory · writesTeam-scoped long-term memory (add, search, get, delete).Persist or retrieve facts for this team/user across sessions.World knowledge, one-off page content (use fetch/search), or global encyclopedic answers.
smart_pipePipeline orchestrator · writesRun a multi-step pipeline (built-in template or custom steps).Several tools in sequence (e.g. research, read, remember). Templates: research, read, remember.A single step — call the specific smart_* tool directly.

What each plan allows

Request rates are the catalogue values enforced by the gateway (per API key, plus the team ceiling); token allowances and log retention are the published plan figures. Compare them side by side on Pricing.

MCP request rate, token allowance and log retention by plan
PlanToken allowanceMCP requests / min / keyTeam ceilingLog retention
Free$02,000,000 tokens / month3030 / min7 days
Profrom $18 / month20,000,000 tokens / month300600 / min30 days
Teamsfrom $55 / month100,000,000 token pool / month6003,000 / min90 days
Enterprisecustom200,000,000+ tokens / month1,2009,999 / minextended, contract-defined

Read from the gateway's own tool metadata and plan catalogue on 2026-10-09. If a number here disagrees with the runtime, the runtime is right — tell us and we will correct the page.

Where to go next

New to the category? Start withAboutfor the is / is-not story, thenDocsfor endpoint, tools, and connect guides. Compare plans onPricing. ExploreSolutions by rolefor teams, agent builders, and individuals, or browse theIndustriestopic library for deeper MCP, RAG, and governance guides. Integrations for the tools you already use live underIntegrations.

When you are ready to try the gateway path: sign in, open Playground, and run a single-tool or pipe preset under plan caps—the same primitives your MCP client will call in production.

Related roles and scenes

Same gateway, different jobs — pick the path that matches how you work.

Related industries

Deeper guides on the same themes — from our industry library.

Browse all industries.