Hosted server

AI-ecosystem data over MCP.

Agents plan against stale knowledge: model prices move, registries grow, tools ship nightly. This server answers from datasets collected from public machine-readable sources — every record carries source_url and collected_at.

Endpoint

Transport	Streamable HTTP (stateless, JSON responses)
URL	POST /mcp
Auth	Authorization: Bearer tg_… (or X-Api-Key)
Protocol	2025-06-18 · 2025-03-26 · 2024-11-05
Free without a key	initialize · tools/list · resources/list · ping · ai_ecosystem_overview

Tools

tool	arguments	credits
ai_ecosystem_overview	—	0
llm_pricing	provider? model? search? min_context? max_input_price? sort? limit?	1
llm_price_compare	models[2..10] input_tokens output_tokens calls?	1
mcp_servers_search	query transport? limit?	1
mcp_server_info	name	1
ai_changelog	tool? limit? include_prereleases?	1

Bulk dumps are MCP resources (tollgate://datasets/llm-pricing, …/mcp-servers, …/changelogs) at 5 credits per read.

Connect a client

Claude Desktop / stdio-only clients

claude_desktop_config.json — via mcp-remote

{
  "mcpServers": {
    "tollgate": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://your-tollgate.dev/mcp",
               "--header", "Authorization: Bearer tg_…"]
    }
  }
}

MCP Inspector

terminal

$ npx @modelcontextprotocol/inspector
# transport: Streamable HTTP · URL: …/mcp · header: Authorization: Bearer tg_…

Official TypeScript client

client.ts

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";

const transport = new StreamableHTTPClientTransport(new URL("https://your-tollgate.dev/mcp"), {
  requestInit: { headers: { Authorization: "Bearer tg_…" } },
});
const client = new Client({ name: "my-agent", version: "1.0.0" });
await client.connect(transport);
const prices = await client.callTool({
  name: "llm_pricing",
  arguments: { provider: "anthropic", model: "haiku" },
});

Datasets & freshness

dataset	source	refresh
llm-pricing	models.dev (vendor list prices), cross-checked against the OpenRouter API; disagreements kept, both numbers shown	npm run collect-data
mcp-servers	registry.modelcontextprotocol.io — latest active versions	npm run collect-data
changelogs	GitHub Releases: claude-code, codex, gemini-cli, ollama, zed, aider, MCP spec	npm run collect-data

Honesty rules: collectors fail loudly when a source is down — the build never invents a row. Account-tier rate limits (RPM/TPM) are not in the dataset yet; model context/output limits are. Tools that don't publish GitHub releases (e.g. Cursor) aren't covered in this snapshot.

REST mirror

The dashboard endpoints double as a tiny REST API: POST /v1/signup, GET /v1/usage, POST /v1/keys/rotate, POST /v1/billing/checkout, GET /v1/public/stats. Same bearer key.

← Error codes