Gram by Speakeasy
Serverless platform to build, curate, and host tools as MCP servers
Turn API specifications into docs, typed SDKs, and an MCP server
Cortex is a neat piece of leverage: if your API already has a decent spec, you get docs, SDKs, and an MCP server from one pipeline instead of maintaining three. The MCP generation is the reason to look at it now, since hand-writing an MCP server per API is exactly the kind of repetitive work a generator should absorb. Output quality depends entirely on spec quality, so teams with thin or drifting specs will get thin results. It is MIT licensed with no hosted service, which means no vendor risk but also no support contract.
Cortex is an MIT-licensed CLI that turns API specifications and Markdown into self-hostable interactive documentation, typed SDKs in eleven languages, and an MCP server exposing each operation as a typed tool.
Cortex takes the position that an API specification already contains everything needed to serve three audiences at once: humans reading documentation, developers importing an SDK, and agents calling tools. From one spec plus Markdown, it generates a documentation site, typed client libraries, and an MCP server where every operation becomes a typed tool an agent can call. The generator accepts OpenAPI, AsyncAPI, GraphQL, gRPC, and OpenRPC, and emits SDKs across TypeScript, Python, Go, Java, Kotlin, Ruby, PHP, C sharp, Rust, C plus plus, and C. Output is deployable anywhere Node runs, so there is no hosted dependency and no per-seat cost. The MCP output is the notable part. Most documentation generators stop at humans; Cortex treats agent consumption as a first-class target, which matters as more API traffic arrives from agents rather than hand-written code. The project is MIT licensed with a public repository and a live demo instance.
Cortex generates self-hostable API documentation, typed SDKs, and an MCP server from a single API specification, under MIT licence.
Cortex is an open-source project published under the cortex-docs organisation, launched to Product Hunt in September 2026 where it placed second in its category.
Its thesis is that one API specification should serve humans, developers, and agents, so documentation, SDKs, and MCP tooling are generated from the same source.
Cortex is a CLI configured via cortex.config.yml that ingests OpenAPI, AsyncAPI, GraphQL, gRPC, and OpenRPC and emits interactive documentation, typed SDKs in eleven languages, and an MCP server where each operation is a typed tool.
All output is self-hostable and deployable anywhere Node runs. The project is MIT licensed with a public demo instance.
API and platform teams with maintained specifications who want documentation, client SDKs, and agent-accessible MCP tooling without maintaining three pipelines.
Developers consuming generated SDKs and docs.
Platform and developer experience teams.
MCP and API tooling communities.
An API team with a clean specification that wants to expose the same API to humans, SDK users, and agents.
Not stated; this is an open-source project rather than a funded company.
OpenAPI, AsyncAPI, GraphQL, gRPC, and OpenRPC.
TypeScript, Python, Go, Java, Kotlin, Ruby, PHP, C sharp, Rust, C plus plus, and C.
Yes. Each API operation is exposed as a typed MCP tool that agent clients can call.
No. It is a CLI that generates output you deploy yourself, anywhere Node runs.
MIT.
Side-by-side pages for pricing, features, and best-fit use cases.
Serverless platform to build, curate, and host tools as MCP servers
Local-first memory for AI coding agents, via one MCP service
Open-source framework for building agents on the Model Context Protocol
Open-source tool-calling platform connecting 600+ tools to AI agents