Skip to main content
Home

Use AI to explore Bittensor

Point any agent at metagraphed — over MCP, a typed SDK, or plain HTTP — and it can find, explain, and call the right Bittensor subnet for a task. No key, no account.
128
Subnets covered
2,271
Callable services
206
MCP tools

01Hand off context

Copy once, ingest once — everything an agent needs to start using metagraphed on its own.

One-file context bundle

Hand this to an agent and it can use metagraphed.

Paste-ready system prompt that turns any agent (Claude, Cursor, …) into a metagraphed-powered Bittensor integration agent.

02Connect your client

MCP is the fastest path — no install in the consuming project. The SDK, skill, and chat links below are alternates for hosts that can't speak MCP.

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

Terminal command
claude mcp add --transport http metagraphed https://api.metagraph.sh/mcp
api.metagraph.sh/mcpserver card206 tools over streamable-http
search_subnetslist_subnetsfind_subnets_by_capabilityget_subnetget_subnet_detailget_subnet_snapshotget_network_healthget_health_historyget_subnet_healthget_subnet_health_trendsget_health_trendsget_subnet_health_percentilesget_subnet_health_incidentsget_subnet_economicsget_subnet_stake_quoteget_stake_action_previewget_economicsget_subnet_trajectoryget_economics_trendsget_subnet_concentrationget_subnet_performanceget_subnet_idle_stakeget_chain_concentrationget_chain_performance
pip install metagraphedPython · metagraphed
npm i @jsonbored/metagraphedTypeScript · @jsonbored/metagraphed
gh skill install JSONbored/metagraphed bittensorBittensor skill

Paste-ready system prompt that turns any agent (Claude, Cursor, …) into a metagraphed-powered Bittensor integration agent.

Open in Claude Open in ChatGPT

03Query the registry live

Query the registry live

Grounded answers and vector search over all 128 subnets — the same data the MCP's 204 tools and 2,292 callable services are built on. Run a real query below.

This calls the live HTTP API directly, over the network from your browser — it proves the registry is reachable and answering, not that a specific MCP client's handshake works. The response time shown is round-trip from here, not from an MCP session.

Same call as

04Try your first prompt

Paste one straight into the client you just connected — each one is a real task, not a demo.

“Evaluate subnet 7 before I stake into it — health, economics, and stake concentration.”

A read on whether a subnet is healthy and how concentrated its stake is, before you commit.

“How is validator hotkey 5FHneW... doing? Current standing, 30-day trend, and who's staking to it.”

A validator's current standing and trend, plus its nominator breakdown.

“I need a subnet that does image generation over an HTTP API — find one and show me how to call it.”

A specific callable service matched to your task, with the actual request to make.

05Deeper integrations

Context files, the OpenAPI contract, GraphQL, bulk data, and everything else the registry exposes directly.

Agent workflows

Copyable REST/npm/Python/MCP call examples, one per workflow.

/agent-workflows.md

llms.txt

The llms.txt convention — a directory of every doc page as plain links.

/llms.txt

llms-full.txt

The full corpus concatenated into one file, for agents that ingest in bulk.

/llms-full.txt

OpenAPI 3.1 contract

Every route, typed. Point any OpenAPI codegen at it for a typed client.

/metagraph/openapi.json

Agent capability catalog

Every callable service in the registry, in one paginated list.

/api/v1/agent-catalog

Coverage depth scorecard

How complete each subnet's registry entry is, scored.

/api/v1/coverage-depth

Semantic search

Vector search over subnets and surfaces — the REST form of Search above.

/api/v1/search/semantic?q=

Ask (grounded Q&A)

Grounded Q&A with citations — the REST form of Ask above.

/api/v1/ask

GraphQL (shaped registry queries)

Shaped queries over the registry instead of stitching multiple REST calls.

/api/v1/graphql

Live request/response fixtures

Real request/response pairs captured live, for writing tests against.

/api/v1/fixtures

Cross-network lineage

Cross-network lineage — which mainnet subnet a testnet one mirrors.

/api/v1/lineage

Bulk CSV datasets

The whole registry as flat CSV, for a notebook or a spreadsheet.

/datasets/index.json

06Task-oriented playbooks

Executed, tested tool-call sequences for the tasks people actually bring — also registered as MCP prompts for harnesses that surface them natively.

Evaluate a subnet before staking

Health, economics, and stake concentration before staking into it.

Monitor my validator

Current standing, 30-day trend, and who's staking to it.

Find a subnet for my app

A plain-language task to a callable subnet's base URL.

Audit an account's history

Reconstruct what one SS58 address has actually done on-chain.