Granary by Speakeasy: The "Utility" of the AI Agent Era

2026-02-13 | Product Hunt | Official Site | GitHub

Interface Breakdown: This is the Gram Playground. On the left is a toolset (13 tools) auto-generated from OpenAPI docs. On the right is an AI chat window (running Claude 3.5 Sonnet), where developers can test how the AI Agent calls API tools in real-time. Simply put, it's a "try-before-you-deploy" sandbox.

30-Second Quick Take

What it does: It turns your existing APIs (OpenAPI docs) into an MCP Server with one click, allowing AI tools like Claude and Cursor to call your APIs directly. In plain English—it gives your AI Agent "hands" to operate your system.

Is it worth watching?: Absolutely. MCP has become the de facto standard for AI Agents to connect to the outside world (used by OpenAI, Google, and Microsoft), and Gram is the most mature "API to MCP" solution available. If you're building AI Agent products, this is a must-know.

Three Key Questions

Does this matter to me?

Target Audience:

• SaaS companies with REST APIs who want to make their APIs "AI-ready."
• Dev teams building AI Agents who need to connect to various external APIs.
• Enterprise IT looking to centrally manage which tools Agents can access.

Is this you? You are the target if:

• You have an API with OpenAPI docs and want an AI to be able to call it.
• You use Claude Desktop / Cursor / OpenAI Agents SDK and want to plug in more tools.
• You're building an AI product that needs to operate third-party systems (Payments, CRM, Databases, etc.).

Use Cases:

• You have an API with 600 endpoints and want to curate 15 specific ones for an AI --> Use Gram.
• You need an AI Agent to help customers configure billing --> Use Gram.
• You want to connect an internal AI assistant to Jira/Slack/Your own system --> Check Composio first; use Gram for highly custom needs.
• You just use Claude to write code --> You don't need this.

What's the payoff?

Is it any good?

The "Aha!" Moments:

• One-Click Generation: Upload docs, and 600 endpoints are automatically curated into 5-30 AI-friendly tools. This curation is the biggest selling point.
• Plug and Play: The generated MCP Server works in Claude Desktop or Cursor with zero configuration.
• Playground Testing: See how the AI calls your tools before you deploy, allowing you to catch errors early.

User Quote:

"I considered building my own MCP Server for five minutes before realizing it would be a productivity black hole." — Founder of Polar

Real Feedback:

Positive: "Working with the Speakeasy team has been great; they are incredibly responsive. We turned our payment API into a production-grade MCP Server in less than a week." — Polar Team Neutral: Nordic APIs noted the setup is "relatively simple, but requires navigating several screens and some time to find documentation."

For Developers

Tech Stack

• Frontend / Dashboard: TypeScript
• Backend: Golang + Goa (design-first API framework)
• Workflow Engine: Temporal
• LLM Gateway: OpenRouter
• Billing: Polar (usage-based)
• Tool Definition: TypeScript SDK + Zod schema validation
• SDK Generation: Speakeasy’s own SDK generator

How it Works

Gram's core logic follows two paths:

Path 1 — OpenAPI Auto-gen: You upload OpenAPI docs, Gram parses every endpoint, and wraps them into MCP-compatible tool definitions. Crucially, it’s not a simple 1:1 mapping; it performs "tool curation"—distilling hundreds of endpoints into a few dozen semantically meaningful high-level tools.

Path 2 — Gram Functions: Write custom tools in TypeScript to chain multiple low-level API calls into one high-level tool (e.g., a "Create Order" tool that calls 5 APIs internally). Start quickly with npm create @gram-ai/function@latest.

Everything is then deployed as a managed MCP Server with OAuth 2.1 + PKCE authentication, ready out of the box.

Open Source Status

• Is it open source?: Yes, fully. github.com/speakeasy-api/gram
• Alternatives: FastMCP (Python ecosystem), openapi-mcp-generator (self-hosted).
• Build-it-yourself difficulty: Medium-High. Writing an MCP Server isn't hard, but reaching Gram's level of auto-curation + OAuth + managed hosting would take 2-3 people about 3 months.

Business Model

• Monetization: SaaS Subscription + Usage-based billing.
• Pricing: Free (1K calls/mo) | Pro ($29/mo for 5K calls) | Enterprise (Custom).
• User Base: Speakeasy as a whole has 3,000+ users (2024 data).

The Big Tech Threat

Cloudflare already offers MCP Server hosting, and Vercel is working on something similar. Gram’s differentiator is "tool curation"—it doesn't just host; it helps you turn a messy API into something an AI can actually understand. This curation layer is something the giants haven't tackled yet. However, if OpenAI or Anthropic releases their own "API to MCP" converter, Gram's moat could become quite shallow.

For Product Managers

Solving the Headache

• The Problem: AI Agents need to call external APIs, but exposing APIs directly to an LLM has three pitfalls: too many tools (LLM confusion), complex auth (OAuth nightmares), and zero control (no visibility into who called what).
• The Pain Level: High. Every AI Agent project connecting to external systems hits this wall. As the Polar founder put it, doing it yourself is a "productivity black hole."

User Personas

• Persona 1: API teams at SaaS companies (10-50 people) needing to make their API accessible to customers' AI Agents.
• Persona 2: Engineers at AI startups needing to quickly integrate multiple third-party systems.
• Persona 3: Enterprise IT admins needing to unify tool permission management for internal Agents.

Feature Breakdown

Competitive Landscape

For Tech Bloggers

The Origin Story

• Sagar Batchu (CEO): Physics major from Harvey Mudd; built a 50+ person engineering team in London for LiveRamp. His family has grown coffee in India for generations—now he's building API tools.
• Simon Yu (COO): Former Product at Flexport and SafeGraph.
• The Mission: Sagar saw how painful API experiences were at big companies and wanted to give everyone a "Stripe-level" experience. Speakeasy started with SDK generation and pivoted hard to being "the MCP company" when the protocol exploded in 2025.

Hot Debates

• Is MCP the "USB-C for AI"? Google’s A2A protocol handles Agent-to-Agent talk, while MCP handles Agent-to-Tool. Are they complementary or competing?
• Who should do the "Tool Curation"? Gram currently relies on manual rules. Can AI eventually do the curation itself?
• The Open Source vs. Managed Conflict: If the code is open, why pay? The answer lies in maintenance, OAuth proxies, and SLAs—but how long will that story hold?

Content Ideas

• "What is MCP? Why is every AI company adopting it?" — Use Gram as the case study for the MCP ecosystem.
• "How to give your AI Agent hands" — A tutorial on using Gram to connect custom APIs.

For Early Adopters

Getting Started

• Time to value: 30 minutes (if you have OpenAPI docs).
• Learning Curve: Low (if you know APIs/MCP) to Medium (if you're learning MCP from scratch).
• Steps:
  1. Sign up at speakeasy.com/product/gram.
  2. Upload your OpenAPI docs.
  3. Curate your toolset in the Dashboard.
  4. Test in the Playground with Claude.
  5. Copy the MCP Server URL into Claude Desktop or Cursor.
  6. For custom logic, use npm create @gram-ai/function@latest.

Watch Out For

1. Early Ecosystem: The toolchain is still maturing. If you hit weird bugs, it might be the protocol, not Gram.
2. Doc Quality is King: If your OpenAPI descriptions are poor, the AI won't know how to use the generated tools.
3. Language Limits: Gram Functions are currently TypeScript-only. Python devs might prefer FastMCP.

For Investors

Market Analysis

• Market Size: Agent orchestration is projected at $30B.
• Adoption: 97M+ monthly SDK downloads; OpenAI, Google, Microsoft, and Anthropic are all in.
• Growth: Gartner predicts 40% of enterprise apps will include AI Agents by late 2026 (up from <5% today).

Timing

• Why now?: MCP was released by Anthropic in Nov 2024 and adopted by all giants by 2025. We are in the prime window for protocol standardization.
• Team: Strong background in API DevEx and scaling engineering teams.
• Funding: ~$26.6M total, led by GV and FPV Ventures.

Conclusion

Gram is the "Heroku" of the MCP era—it’s the connection layer between Agents and the world. The market is huge and the timing is perfect, but their long-term moat depends on whether "tool curation" remains a distinct, defensible value add.

2026-02-13 | Trend-Tracker v7.3 | Data based on WebSearch + X + Gemini Image Analysis