Gram by Speakeasy

What Gram by Speakeasy MCP server does

Gram by Speakeasy MCP server is the open-source stack behind Speakeasy’s AI control plane. It provides a central layer for hosting, governing, and distributing MCP servers, Skills, Plugins, and Assistants across an organization. The platform is designed to connect agents with SaaS vendors, APIs, and internal systems while keeping access and activity under organizational control.

The Gram by Speakeasy MCP server covers four main areas:

• Observe: Track AI usage by token count or cost, review expensive sessions, create budgets, and evaluate tool usage. Telemetry is built on OpenTelemetry.
• Secure: Inspect prompts, responses, and agent actions in real time. Policies can block, redact, or log sensitive data and help detect prompt injection, PII, and secrets.
• Connect: Provide a common connection layer for MCP services, with policy enforcement, granular permissions, private networking, and on-demand tunneling.
• Distribute: Assign MCPs, Skills, Plugins, and Assistants to teams through enterprise roles, with permissions at the team, server, and tool levels.

How it works

Access control uses role-based permissions and OAuth 2.1. The platform can synchronize with enterprise identity providers including Okta, Azure AD, and Google Workspace. Every tool call, permission change, and access event is logged and searchable according to the project description.

Usage data can be examined through the platform’s MCP interface and its built-in assistant. Cost and token views help teams investigate sessions and measure tool effectiveness. The repository also describes support for Claude, Claude Code, ChatGPT, Codex, Gemini, Cursor, and GitHub Copilot.

Setup and configuration

There are two documented paths. Hosted use starts by signing up through Speakeasy’s Gram application and following the platform quickstart. Local development uses the repository’s ./zero script, which installs dependencies, runs pending database migrations, and starts the dashboard and services.

Local setup depends mainly on Mise and Docker. The script can guide installation when either is missing. After the stack starts, mise seed creates a local organization with sample APIs, agent sessions, telemetry, risk findings, teammates, and an API key. Running the seed command again resets that seeded data. Separate Git worktrees can run complete stacks on different ports for parallel development.

Tools and capabilities

The Gram by Speakeasy MCP server supports centralized management of MCP resources and related agent components. Its documented capabilities include:

• Hosting and distributing MCP servers, Skills, Plugins, and Assistants.
• Enforcing team-, server-, and tool-level permissions.
• Applying policies to prompts, responses, and agent actions.
• Redacting, blocking, or recording sensitive information before it leaves an environment.
• Recording access events, tool calls, and permission changes.
• Measuring usage with token and cost data.
• Connecting private systems through networking and tunneling options.
• Exporting and interacting with OpenTelemetry-based usage data.

Limitations and notes

The Gram by Speakeasy MCP server is an organization-level control plane, not a description of one application-specific MCP toolset. The supplied material does not list individual MCP tool names or document environment variables for deployment. It also does not provide detailed client configuration examples.

Running the complete project locally requires Mise, Docker, the repository’s startup script, and its supporting services. Hosted access and local development are described separately. Security certifications, including SOC 2 Type II and ISO 27001, are stated for the platform in the supplied project material; deployment-specific compliance should be evaluated independently.