How does it work?
Step 1: Load your API
Section titled “Step 1: Load your API”Start by uploading your OpenAPI documents into Gram. It will parse these and generate a list of tool definitions it found. You can load multiple documents representing different services in your organization. It all becomes a sea of tools for you to work with.
Step 2: Create toolsets
Section titled “Step 2: Create toolsets”Your API may have tens or hundreds of endpoints that get mapped to tool definitions. Exposing all of these to an LLM is a bad idea in most cases because it can lead to tool confusion. Instead, you want to arrange tools into sets (we call them toolsets) that relate to specific tasks and workflows.
Step 3: Fill out auth and server details
Section titled “Step 3: Fill out auth and server details”APIs often require authentication/authorization details as well information on where they are hosted (the server URL). In Gram, these concepts are combined into environments. Before you can start interacting with tools you want to fill out credentials and server URLs. We create a “default” environment for you that you can start filling out after uploading your OpenAPI document.
Step 4: Test your toolsets out
Section titled “Step 4: Test your toolsets out”Use the playground on the Gram dashboard to test your toolsets out. You’ll have access to popular language models. With your chosen toolset, chat with the LLM and evaluate how well it can use the tools you just created for a given task.
Step 5: Integrate!
Section titled “Step 5: Integrate!”Hosted MCP server
Section titled “Hosted MCP server”Every Gram toolset you create is exposed as a hosted MCP server. Model Context Protocol (MCP) has become a de-facto standard for packaging tools and making them available to various language models and LLM clients. On the Gram dashboard, navigate to a toolset and you’ll find a section guiding you on how to add the MCP server to your favorite LLM client.
Agentic frameworks and SDKs
Section titled “Agentic frameworks and SDKs”You’ve got everything in place to start integrating with popular agentic frameworks such as Langchain, OpenAI Agents SDK and Vercel AI SDK. We’ve built a Gram Python SDK and TypeScript SDK that help you integrate with all of these frameworks in your Python or TypeScript codebase.
Step 6: Iterate
Section titled “Step 6: Iterate”It’s likely you’ll find that some of your API-endpoints-turned-tools are not being used by LLMs or getting confused with other tools. In Gram, you can modify tool descriptions and names beyond what was chosen from your OpenAPI document. We call these modifications tool variations. Taking the time to prompt-engineer your tools with variations along with effectively cherry-picking them into task-focused toolsets can result in very effective agents down the line.