MCP Transport

MCP (Model Context Protocol) allows any LLM client to interact with a BEST-compliant service directly — discovering commands and queries, reading state, and sending commands — without any bespoke integration.

LLM Client
Copilot / Claude / ChatGPT
any MCP-capable client
MCP tools
best-mcp
MCP Server
stdio or http transport
BEST HTTP
Service
BEST Endpoint
any compliant API

Choosing a Transport

MCP is an adapter for clients you don't control — not the recommended path for code you own. HTTP is the baseline transport, and the BEST HTTP surface is deliberately self-describing: catalogues, schemas and workflows are designed to be consumed by an agent directly, with no adapter in between.

  • Off-the-shelf MCP-capable client (Claude Desktop, VS Code Copilot, Cursor, ChatGPT Desktop, CLI agents): use MCP. It is the only plug-in mechanism those clients offer, and best-mcp gives them the full command/query surface with zero bespoke code.
  • Your own code (a bespoke agent, a backend, tooling you build and deploy): call the BEST HTTP surface directly. Every best-mcp tool is a thin wrapper over exactly one HTTP endpoint, so placing the MCP server between your own client and the service adds a network hop and a deployment to operate, flattens structured BEST error responses (HTTP status + error code) into prose strings, and widens your supply chain — while providing nothing a small HTTP client in your own codebase wouldn't.

Rule of thumb: never put best-mcp between a first-party client and a BEST service. Declaring the mcp transport in your manifest is for the consumers you don't ship code to.

Mapping

BEST Concept MCP Concept
Connection list (multi-connection mode) MCP tool: list_connections
Command catalogue (GET /commands) MCP tool: get_command_catalogue
Command schema (GET /commands/{schema}/{version}) MCP tool: get_command_schema
Command ingestion (POST /commands) MCP tool: send_command
Query catalogue (GET /queries) MCP tool: get_query_catalogue
Query schema (GET /queries/{schema}/{version}) MCP tool: get_query_schema
Query execution (GET /queries/{schema}) MCP tool: execute_query
Push event delivery MCP server-to-client notifications

Result

Any LLM client (ChatGPT Desktop, GitHub Copilot, Claude Desktop, Cursor) becomes a capable caller of any BEST-compliant service — with full command and query discovery, no hardcoded integration.

Reference Implementation — best-mcp

best-mcp is the reference MCP server for BEST. It is generic — it works with any BEST-compliant endpoint. Point it at any BEST HTTP surface and it exposes the full command and query surface as MCP tools.

npx @behavioralstate/best-mcp

Pin a version in production. Unversioned npx @behavioralstate/best-mcp resolves latest from npm at start-up — convenient on a workstation, but a server deployment then silently picks up new code (including any compromise of the npm package) on its next cold start, with your callers' credentials flowing through it. Production deployments should pin an exact version (e.g. npx @behavioralstate/best-mcp@1.7.0) and upgrade deliberately.

Configuration

best-mcp supports three configuration modes, checked in priority order.


Mode 1 — Per-app env vars (recommended)

One set of BEST_<APP>_* variables per application. The app name must be a single uppercase word (letters and digits only, e.g. TRADING, HR).

Required:

Variable Description
BEST_<APP>_BASE_URL Root URL of the BEST HTTP surface
BEST_<APP>_API_KEY Credential — not required when AUTH_TYPE=none

Optional:

Variable Default Description
BEST_<APP>_TENANT_ID When set, auto-generates two named connections: <app>/tenantBASE_URL/tenants/TENANT_ID and <app>/platformBASE_URL. When omitted, generates one connection: <app>.
BEST_<APP>_AUTH_TYPE apikey* bearer · apikey · nonedefaults to apikey in Mode 1 (Modes 2 & 3 default to bearer)
BEST_<APP>_AUTH_HEADER X-Api-Key Header name when AUTH_TYPE=apikey and AUTH_IN=header
BEST_<APP>_AUTH_IN header header or query — where the key is sent when AUTH_TYPE=apikey
BEST_<APP>_AUTH_PARAM apikey Query parameter name when AUTH_IN=query

Auth types:

Default differs by mode. Mode 1 defaults to apikey because BEST services typically use API key headers. Modes 2 and 3 default to bearer for backward compatibility.

AUTH_TYPE Credential transport Extra vars
apikey (Mode 1 default) Custom header (default X-Api-Key) or query param AUTH_HEADER or AUTH_IN=query + AUTH_PARAM
bearer (Modes 2 & 3 default) Authorization: Bearer <key> none
none No credentials API_KEY not required

Example — admin with tenant + platform surfaces (generates two connections from one config block):

BEST_TRADING_BASE_URL=https://api.example.com/best
BEST_TRADING_API_KEY=your-api-key
BEST_TRADING_TENANT_ID=your-tenant-id
BEST_TRADING_AUTH_TYPE=apikey

Connections produced: trading/tenant and trading/platform.


Mode 2 — BEST_CONNECTIONS JSON array

For cases where per-app vars are not flexible enough. Set BEST_CONNECTIONS to a JSON array of fully-explicit connection objects (name, endpoint, apiKey, authType, authHeader, authIn, authParam, description).


Mode 3 — Legacy single connection

For simple single-endpoint setups. Set BEST_ENDPOINT, BEST_API_KEY, and optionally BEST_AUTH_TYPE, BEST_AUTH_HEADER, BEST_AUTH_IN, BEST_AUTH_PARAM.


Transport vars (all modes):

Variable Default Description
MCP_TRANSPORT stdio stdio or http
MCP_HTTP_PORT 3000 HTTP port when MCP_TRANSPORT=http

stdio — VS Code Copilot, Cursor, Claude Desktop

{
  "mcpServers": {
    "best": {
      "command": "npx",
      "args": ["@behavioralstate/best-mcp"],
      "env": {
        "BEST_TRADING_BASE_URL": "https://api.example.com/best",
        "BEST_TRADING_API_KEY": "<your-api-key>",
        "BEST_TRADING_TENANT_ID": "<your-tenant-id>",
        "BEST_TRADING_AUTH_TYPE": "apikey"
      }
    }
  }
}

For multiple apps, add more BEST_<APP>_* variable groups to the same env block.

HTTP — ChatGPT Desktop

Start in HTTP mode and expose via a tunnel:

MCP_TRANSPORT=http MCP_HTTP_PORT=3001 \
  BEST_ENDPOINT=https://api.example.com/BEST \
  BEST_API_KEY=<key> \
  npx @behavioralstate/best-mcp

ngrok http 3001

Then in ChatGPT Desktop: Settings → Apps & Connectors → Create, connector URL: https://<subdomain>.ngrok.app/mcp

Intended LLM flow

list_connections (multi-connection only) → identify and confirm the target connection

get_command_catalogue          → discover what commands this service accepts
get_command_schema             → learn the exact fields and the required source value
send_command                   → send the command (CloudEvent envelope built automatically)

get_query_catalogue            → discover available read queries
get_query_schema               → learn parameters and response shape
execute_query                  → read current state synchronously

CloudEvent construction

send_command builds the CloudEvent 1.0 envelope automatically:

  • type is derived from the schema name via PascalCase conversion (configure-broker → ConfigureBroker)
  • dataschema is set to the absolute catalogue URI {endpoint}/commands/{schema}/{version} (e.g. https://api.example.com/best/commands/configure-broker/1.0)
  • source must be supplied by the caller — the required value is documented in the schema description returned by get_command_schema; never invent or default it

Transport Configuration

The mcp block in the service definition declares how to reach the MCP server:

"mcp": {
  "transport": "http",
  "server": "https://mcp.example.com/mcp"
}
Field Type Required Description
transport string yes MCP transport type: "stdio", "sse", or "http"
server string yes MCP server identifier or URL
push boolean no When true, the server supports server-to-client push notifications for domain events. Callers should prefer this channel over polling GET /events.
authentication object no Authentication requirements for connecting to this MCP server

MCP transport is optional — HTTP is the baseline. MCP is declared in the /.well-known/best manifest only if the endpoint supports it.

Authentication

If the MCP server requires authentication, it declares this in an authentication block on the mcp transport object. Consumers — including AI agents and IDE tooling such as VS Code Copilot — must read this block to know what credentials to supply when connecting.

"mcp": {
  "transport": "http",
  "server": "https://mcp.example.com/mcp",
  "authentication": {
    "type": "apiKey",
    "headers": [
      { "name": "X-Api-Key",   "description": "Your API key" },
      { "name": "X-Tenant-Id", "description": "Your tenant identifier", "example": "acme" }
    ],
    "docs": "https://docs.example.com/authentication"
  }
}

Authentication Fields

Field Type Required Description
type string yes One of: "none", "bearer", "apiKey", "oauth2"
headers array no Required headers for apiKey type — use when more than one header is needed (e.g. both an API key and a tenant ID). For single-header API key auth, headers with one entry is preferred over scheme + in.
scheme string no For bearer: the Authorization header prefix (e.g. "Bearer")
tokenUrl string no Token endpoint URL for oauth2 or token-based flows
scopes string[] no Required OAuth2 / token scopes
docs string no URL to human-readable authentication documentation

Header Descriptor

Each entry in headers describes one required HTTP header:

Field Type Required Description
name string yes HTTP header name (e.g. "X-Api-Key")
description string no Human-readable description of what value to supply
example string no Pre-filled example value. Tooling may use this to populate the header automatically — for instance, a per-tenant manifest may pre-fill the resolved tenant ID here so VS Code can generate a complete MCP server config without manual input.

Auth Type Examples

API key — single header:

"authentication": {
  "type": "apiKey",
  "headers": [
    { "name": "X-Api-Key", "description": "Your API key" }
  ],
  "docs": "https://docs.example.com/auth"
}

API key — multiple headers (e.g. key + tenant ID):

"authentication": {
  "type": "apiKey",
  "headers": [
    { "name": "X-Api-Key",   "description": "Your API key" },
    { "name": "X-Tenant-Id", "description": "Your tenant identifier", "example": "acme" }
  ],
  "docs": "https://docs.example.com/auth"
}

Bearer token:

"authentication": {
  "type": "bearer",
  "scheme": "Bearer",
  "docs": "https://docs.example.com/auth"
}

OAuth2:

"authentication": {
  "type": "oauth2",
  "tokenUrl": "https://auth.example.com/oauth2/token",
  "scopes": ["mcp:read", "mcp:write"],
  "docs": "https://docs.example.com/auth"
}

Tooling hint: The example field on a header is intended for IDE tooling (e.g. VS Code Copilot's MCP server config). When consuming a per-tenant manifest, example values may be pre-filled so tooling can generate a ready-to-use MCP server config with no manual entry required.

MCP authentication vs. root authentication. The root authentication block in the manifest describes credentials for the HTTP API. The mcp.authentication block describes credentials for the MCP server specifically. These may use the same mechanism or different ones — each transport declares its own requirements independently.

Push Event Delivery

When a caller maintains an active MCP session and "push": true is declared on the mcp block, the server may push domain events to the caller using MCP's server-to-client notification mechanism. Events are delivered as MCP notifications matched by the correlation identifier of a previously submitted command.

"mcp": {
  "transport": "http",
  "server": "https://mcp.example.com/mcp",
  "push": true
}

When "push": true is present, callers should prefer this channel over polling GET /events. The io.best.agents.events capability in the manifest declares "push": { "mcp": true } when this channel is active — see Discovery.

spec/v0.9.1