Agent Manifest Protocol (AMP)

What Agent Manifest Protocol (AMP) MCP server does

The Agent Manifest Protocol (AMP) MCP server exposes AMP’s manifest-oriented API discovery functions through Model Context Protocol. It is intended for agents that need to inspect an API’s declared capabilities, usage guidance, pricing, authentication, reliability information, and endpoints before deciding how to use that API.

The server focuses on three operations described by the project: retrieving an AMP manifest, checking a manifest against the protocol’s JSON Schema, and testing an action against constraints declared by the manifest. These operations help an agent work from provider-supplied metadata instead of relying only on hardcoded API knowledge.

AMP manifests are served by APIs at /.well-known/agent-manifest.json. A manifest can describe the API’s endpoints, endpoint parameters, response semantics, pricing model, authentication requirements, reliability details, and notes written for agents.

How it works

The Agent Manifest Protocol (AMP) MCP server acts as an MCP-facing layer over AMP manifest data. An MCP client sends a tool request through the configured transport, and the server performs the corresponding manifest fetch, schema validation, or action-constraint check.

For local development, the package provides a stdio process. The README identifies this mode for Cursor and Claude. A separate HTTP package starts an SSE endpoint for a hosted hub, with port 8787 used by default unless PORT is set differently.

Manifest validation uses the AMP JSON Schema rather than only checking whether a document can be parsed as JSON. Action checks use the declarations in the retrieved manifest, so the result depends on what the API provider has described in that document.

Setup and configuration

Run the stdio server with:

npx @agentmanifest/mcp-server

Run the SSE HTTP variant with:

npx @agentmanifest/mcp-server-http

The HTTP process listens on the value of PORT; the documented default is 8787. The provided material does not specify additional required environment variables, credentials, or configuration files for the MCP packages.

The repository’s broader development setup requires Node.js 18 or later. PostgreSQL and Docker are documented for running the wider validator and registry project, but they are not identified as prerequisites for the standalone MCP commands.

Tools and capabilities

The Agent Manifest Protocol (AMP) MCP server supports these documented capabilities:

• Fetch an AMP manifest from an API’s well-known manifest location.
• Validate manifest JSON against the AMP JSON Schema.
• Check an action against constraints declared by a manifest.
• Serve MCP over stdio for local clients.
• Serve MCP over SSE HTTP for a hosted hub.

The server is concerned with manifest discovery and verification. The supplied material does not document tools for calling arbitrary API endpoints, submitting registry listings, or handling payments.

Limitations and notes

Results depend on the availability and accuracy of an API’s manifest. Validation can identify whether a manifest conforms to the schema, but the README does not claim that it independently verifies the provider’s data quality, uptime, pricing, or endpoint behavior.

The project distinguishes the MCP server from the wider AMP protocol, validator, and registry. The repository also describes a separate validator CLI, web API, and public registry; those components should not be assumed to be exposed as MCP tools unless specifically documented for the MCP package.

The specification is licensed CC0 1.0, while the project code is licensed MIT.