MCP

Bittensor in a box — the whole registry as Model Context Protocol tools over one streamable-HTTP endpoint, with a curated 23-tool core profile for context-limited agents. No API key required.

Point any MCP client at the registry and it gets metagraphed as callable tools: subnet discovery and comparison, per-subnet health and economics, OpenAPI schemas for every registered subnet API (callable through the registry itself), chain blocks/extrinsics/events, account lookup, and grounded question-answering — the same data every REST route serves, shaped for agents. Bittensor in a box: one connect, no key, no account, no local install.

https://api.metagraph.sh/mcp        streamable HTTP — full catalog, 240 tools
https://api.metagraph.sh/mcp/core   streamable HTTP — core profile, 23 tools

Which endpoint

Default to /mcp/core. The full /mcp tool list serializes to ~406K tokens of tool schemas — more than most model context windows — and a client that holds tool definitions in context pays that on every session. /mcp/core lists the curated golden path (discover → verify → integrate → call → screen economically) at ~43K tokens.

	`/mcp`	`/mcp/core`
Tools listed	240	23 (the curated golden path)
`tools/list` context	~406K tokens	~43K tokens
Tools callable	all 240	all 240 — listing filters, calls don't
Pick it when	context is cheap (subagents, offline tooling)	context is model context — most agents

A context diet, not an authorization boundary

A session connected to /mcp/core can still tools/call every one of the 240 tools. When a thin session needs more, get_more_tools pages additional tool definitions in on demand, and ask answers whole questions without any further tools at all — a core session is never stranded.

Authentication

Optional. Anonymous access works on both endpoints at 100 requests/60s per client IP. Clients that support MCP authorization prompt an OAuth sign-in (GitHub) on first connect; an Authorization: Bearer mg_… API key header works too. Authenticating raises rate limits (500 requests/60s per account, higher on paid tiers) — it does not unlock additional tools or surfaces.

Connect a client

Claude Code

One CLI command, no config file to hand-edit:

claude mcp add --transport http metagraphed https://api.metagraph.sh/mcp/core

Claude Desktop & claude.ai

Both take the endpoint as a custom connector: Settings → Connectors → Add custom connector, then paste https://api.metagraph.sh/mcp/core. Claude Desktop also accepts a config-file entry:

claude_desktop_config.json

{
  "mcpServers": {
    "metagraphed": {
      "transport": { "type": "http", "url": "https://api.metagraph.sh/mcp/core" }
    }
  }
}

Cursor

.cursor/mcp.json

{
  "mcpServers": {
    "metagraphed": { "url": "https://api.metagraph.sh/mcp/core" }
  }
}

VS Code

.vscode/mcp.json

{
  "servers": {
    "metagraphed": { "type": "http", "url": "https://api.metagraph.sh/mcp/core" }
  }
}

Any other client

Any client that speaks streamable-HTTP MCP connects directly — no install, no key:

Generic streamable-HTTP entry

{
  "type": "streamable-http",
  "url": "https://api.metagraph.sh/mcp/core"
}

First calls to try

All three are in the core profile:

Tool	Try	What it shows
`search_subnets`	`{"query": "inference"}`	Keyword discovery across every subnet's curated profile.
`ask`	`{"question": "which subnets expose an inference API?"}`	A grounded natural-language answer with citations — the whole-question shortcut.
`call_subnet_surface`	a surface id from `list_subnet_apis`	Calling a subnet's own registered API through the registry, schema-checked.

Every tool mirrors a documented REST route — the API reference covers the request/response shapes behind them, and For agents has the rest of the machine surface: playbooks, llms.txt, the served skill, and prompt-ready links.

On this page