@flarewatch/mcp

@flarewatch/mcp

Bring the Flare network into your AI assistant.
Validator scores, FTSO providers, prices, your delegations — all read-only, all attributed.

What this is

A thin stdio↔HTTPS bridge for the FlareWatch MCP server at https://mcp.flarewatch.io/api/mcp. Lets stdio-only MCP clients (Claude Desktop, Cline, older Cursor / Continue builds) reach the remote endpoint without each client having to implement the Streamable HTTP transport.

If your client speaks Streamable HTTP natively (recent Cursor, Continue), you don't need this bridge — point your client at the URL directly.

Why use FlareWatch's MCP

• Real data. Every response is grounded in live on-chain reads (validator weights, FTSO submissions, reward epochs) or cron-refreshed caches — not LLM speculation.
• Attributed. Every response carries source, methodology_uri, data_freshness, and an HMAC verification_signature. AI clients can cite specifically and other AIs can verify.
• Read-only by architecture. 12 independent security walls. The MCP physically cannot mutate state in your wallet or in any FlareWatch system. Even worst-case prompt injection has no path to writes.
• Free + untiered for public tools. No delegation requirement, no per-call quota, no API key.
• Neutral. No "recommended by FlareWatch" anywhere. FlareWatch's own validator appears at its natural rank under the same scoring algorithm as every other validator.

Available tools

Public (no auth — 13 tools)

Tool	What it returns
get_flare_validators	Top-N Flare validators sortable by score / stake / APY
get_validator_score	Full per-dimension breakdown for one NodeID
get_ftso_providers	FTSO data providers on Flare or Songbird
get_ftso_provider_metrics	Detailed metrics for one provider
get_network_status	Current reward-epoch state on either network
get_flarewatch_validator	FlareWatch's own validator stats (same scoring as everyone)
get_flarewatch_sgb_entity	FlareWatch's SGB FTSO entity (live on-chain)
get_flare_prices	FLR / SGB / XRP prices in USD + fiat
get_recent_agent_actions	FlareWatch autonomous agent's action feed
get_agent_capabilities	Agent's autonomy by playbook family
compare_validators	Side-by-side up to 10 validators
compare_ftso_providers	Side-by-side up to 10 providers
verify_response	HMAC-verify a previously-served response

Wallet-scoped (OAuth 2.1 + PKCE — 4 tools)

Tool	What it returns
get_my_delegations	Your WFLR FTSO + P-chain validator delegations
get_my_claimable	Your claimable rewards across networks
get_my_portfolio_summary	Your FLR + SGB balances + USD valuations
get_my_tax_records	Structured tax line items — NOT tax advice

Install

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "flarewatch": {
      "command": "npx",
      "args": ["-y", "@flarewatch/mcp"]
    }
  }
}

Cmd-Q Claude Desktop fully (close window isn't enough), reopen. New conversation → tools icon → confirm flarewatch shows up.

Cursor

Cursor speaks Streamable HTTP natively — skip this bridge:

{
  "url": "https://mcp.flarewatch.io/api/mcp"
}

Continue.dev

.continue/config.json:

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "streamableHttp",
          "url": "https://mcp.flarewatch.io/api/mcp"
        }
      }
    ]
  }
}

Cline

Same as Claude Desktop — Cline's MCP config accepts the same npx -y @flarewatch/mcp invocation.

Any other MCP client with stdio support

npx -y @flarewatch/mcp

The bridge speaks MCP-over-stdio on its standard input/output and forwards every JSON-RPC frame to the remote endpoint over HTTPS.

Raw HTTP test

curl -s -X POST https://mcp.flarewatch.io/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | jq .

Example queries (once attached)

Paste any of these into Claude / Cursor / Continue:

• "Use the FlareWatch MCP to list the top 5 Flare validators by score."
• "What's the current FLR price from the FTSO oracle?"
• "Compare these three validators side by side: [NodeID-A, NodeID-B, NodeID-C]"
• "What's the FTSO reward epoch state on Songbird right now?"
• "Show me my WFLR delegations." (requires OAuth — sign in at flarewatch.io/services/mcp/authorize)

Environment overrides

Variable	Default	Purpose
FLAREWATCH_MCP_URL	https://mcp.flarewatch.io/api/mcp	Endpoint to forward to. Useful for preview deploys or local dev.
FLAREWATCH_MCP_TIMEOUT_MS	30000	Per-request HTTP timeout.

Verification

Every tool response carries a verification_signature HMAC. To verify a citation came from this server:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "verify_response",
    "arguments": {
      "payload": { ... the structured response without verification_signature ... },
      "signature": "the signature you received"
    }
  }
}

Returns { "valid": true } if the signature matches the canonical payload.

What this bridge does NOT do

It's intentionally dumb (~150 LoC). All real logic — tool definitions, rate limiting, output classification, quarantine, audit logging, response signing — lives in the server at mcp.flarewatch.io/api/mcp. The bridge just forwards JSON-RPC frames between stdio and HTTPS.

No caching. No retries. No fallback. If the remote endpoint is down, the bridge surfaces the upstream error rather than attempting recovery.

Links

• Web dashboard: flarewatch.io
• MCP product page: flarewatch.io/services/mcp
• Defense posture (live walls): flarewatch.io/services/mcp/status
• Scoring methodology: flarewatch://methodology/scoring (resources/read)
• Source: github.com/nbridges333/flarewatch
• Companion: Flare protocol docs MCP: dev.flare.network/mcp

License

MIT.