mcpbeat

Shredstream

agiprolabs/shredstream

Pre-execution Solana transaction streaming via Jito ShredStream, Shyft RabbitStream, and Triton Deshred

13k tokens
context cost
the whole folder, loaded on every use
6
files
ships runnable scripts
0
copies elsewhere
how many repositories repackaged it
257
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/agiprolabs/claude-trading-skills --skill shredstream

The instruction itself

20 sections, as written by the author

ShredStream — Pre-Execution Solana Data

ShredStream gives you transaction data before the validator executes the block — typically 100-500ms earlier than standard Yellowstone gRPC. You see transaction *intent*, not confirmed results.

This is the fastest path to Solana data for time-critical trading strategies.

How It Works

Solana validators produce blocks by serializing transactions into shreds (~1,228 bytes each, sized for UDP MTU). Shreds propagate through Turbine (Solana's fanout protocol, 2-3 hops). ShredStream bypasses Turbine by receiving shreds directly from leader validators via Jito's Block Engine.

Leader Validator
     │
     ├── Turbine (standard, 2-3 hops, 200-500ms)
     │       └── Your RPC Node → Yellowstone gRPC (post-execution)
     │
     └── Jito Block Engine (direct)
              └── ShredStream Proxy (your server)
                   ├── UDP shreds → Your RPC/Validator (faster block building)
                   └── gRPC entries → Your Trading Bot (decoded transactions)

What You Get vs. What You Don't

| Available (Pre-Execution) | NOT Available (Needs Execution) |

|---------------------------|--------------------------------|

| Transaction signatures | Success/failure status |

| Account keys (pubkeys) | Balance changes (pre/post) |

| Instructions (program, accounts, data) | Log messages |

| Address lookup table references | Inner instructions (CPI) |

| Slot number | Token balance changes |

| | Compute units consumed |

Key tradeoff: Speed for completeness. You see what's *about to happen* but can't confirm it actually succeeded. Some transactions you see will ultimately fail.

Three Ways to Get Pre-Execution Data

| Provider | Product | Latency | Access | Cost |

|----------|---------|---------|--------|------|

| Jito | ShredStream Proxy | ~10-50ms from leader | Apply + auth keypair | Free (beta) |

| Shyft | RabbitStream | ~15-100ms faster than gRPC | Shyft gRPC plan | From $199/mo |

| Triton | Deshred (SubscribeDeshred) | ~6.3ms p50 from shred | Triton customer | ~$2,900+/mo |

See references/providers_compared.md for detailed comparison.

Option 1: Jito ShredStream Proxy

The most direct approach — run Jito's open-source proxy on your own server.

Get Access

  • Generate a Solana keypair: solana-keygen new -o shred_auth.json
  • Apply at Jito's form with your public key
  • Wait for approval (your keypair gets whitelisted)
  • No staking requirement, free during beta

Run the Proxy

# Clone and build
git clone https://github.com/jito-labs/shredstream-proxy.git --recurse-submodules
cd shredstream-proxy

# Run with gRPC enabled (key flag: --grpc-service-port)
RUST_LOG=info cargo run --release --bin jito-shredstream-proxy -- shredstream \
    --block-engine-url https://mainnet.block-engine.jito.wtf \
    --auth-keypair /path/to/shred_auth.json \
    --desired-regions ny,amsterdam \
    --dest-ip-ports 127.0.0.1:8001 \
    --grpc-service-port 7777

Docker (host networking required for UDP):

docker run -d --name shredstream-proxy --rm \
  --network host \
  -e RUST_LOG=info \
  -e BLOCK_ENGINE_URL=https://mainnet.block-engine.jito.wtf \
  -e AUTH_KEYPAIR=/app/shred_auth.json \
  -e DESIRED_REGIONS=ny,amsterdam \
  -e DEST_IP_PORTS=127.0.0.1:8001 \
  -e GRPC_SERVICE_PORT=7777 \
  -v /path/to/shred_auth.json:/app/shred_auth.json \
  jitolabs/jito-shredstream-proxy shredstream

Configuration

| Parameter | Description | Example |

|-----------|-------------|---------|

| BLOCK_ENGINE_URL | Jito block engine endpoint | https://mainnet.block-engine.jito.wtf |

| AUTH_KEYPAIR | Path to whitelisted Solana keypair | shred_auth.json |

| DESIRED_REGIONS | Max 2, comma-separated | ny,amsterdam |

| DEST_IP_PORTS | Where to forward raw shreds (UDP) | 127.0.0.1:8001 |

| GRPC_SERVICE_PORT | Enable gRPC entry streaming | 7777 |

| SRC_BIND_PORT | Incoming shred UDP port | 20000 |

Available regions: amsterdam, dublin, frankfurt, london, ny, salt-lake-city, singapore, tokyo

Verify It's Working

# Check shreds are arriving via UDP
sudo tcpdump 'udp and dst port 20000'
# Should see many ~1200-byte packets continuously

Consume via gRPC

use jito_protos::shredstream::{
    shredstream_proxy_client::ShredstreamProxyClient,
    SubscribeEntriesRequest,
};

let mut client = ShredstreamProxyClient::connect("http://127.0.0.1:7777").await?;
let mut stream = client
    .subscribe_entries(SubscribeEntriesRequest {})
    .await?
    .into_inner();

while let Some(entry) = stream.message().await? {
    let entries: Vec<solana_entry::entry::Entry> =
        bincode::deserialize(&entry.entries)?;

    for e in &entries {
        for tx in &e.transactions {
            let sig = tx.signatures[0];
            let msg = tx.message();
            // Parse instructions, accounts, etc.
        }
    }

    println!("Slot {}: {} entries, {} transactions",
        entry.slot,
        entries.len(),
        entries.iter().map(|e| e.transactions.len()).sum::<usize>()
    );
}

Option 2: Shyft RabbitStream

Drop-in replacement for Yellowstone gRPC — same SubscribeRequest format, just a different endpoint. Easiest way to get pre-execution data without running infrastructure.

export GRPC_ENDPOINT="https://rabbitstream.ny.shyft.to"
export GRPC_TOKEN="your-shyft-x-token"
# Same code as yellowstone-grpc, just different endpoint
import grpc

endpoint = "rabbitstream.ny.shyft.to"
token = os.environ["GRPC_TOKEN"]

# ... standard Yellowstone connection code ...
# Subscribe to transactions — same filter format
request = SubscribeRequest(
    transactions={
        "pumpfun": SubscribeRequestFilterTransactions(
            account_include=["6EF8rrecthR5Dkzon8Nwu78hRvfCKubJ14M5uBEwF6P"],
            vote=False,
            failed=False,
        )
    },
    commitment=CommitmentLevel.PROCESSED,
)

Limitations: Only transaction filters work. No account, slot, or block subscriptions. The meta field is empty (no execution results).

Regional endpoints: rabbitstream.{ny,va,ams,fra}.shyft.to

Option 3: Triton Deshred

Lowest latency (~6.3ms p50) via Triton's SubscribeDeshred RPC. Same Yellowstone client, different method.

// Requires yellowstone-grpc-client with Deshred support
let (mut tx, mut stream) = client.subscribe_deshred().await?;

tx.send(SubscribeDeshredRequest {
    deshred_transactions: hashmap!{
        "pumpfun".to_string() => SubscribeRequestFilterDeshredTransactions {
            vote: Some(false),
            account_include: vec!["6EF8rrecthR5Dkzon8Nwu78hRvfCKubJ14M5uBEwF6P".into()],
            ..Default::default()
        }
    },
    ..Default::default()
}).await?;

Access: Triton customers only, paid beta, requires their custom Agave validator fork.

Parsing Pre-Execution Transactions

Without execution metadata, parsing is simpler but requires program-specific knowledge.

Identify the Program

# From a raw VersionedTransaction (post-deserialization)
account_keys = [str(k) for k in tx.message.account_keys]
for ix in tx.message.instructions:
    program_id = account_keys[ix.program_id_index]
    if program_id == "6EF8rrecthR5Dkzon8Nwu78hRvfCKubJ14M5uBEwF6P":
        # This is a PumpFun instruction
        discriminator = ix.data[:8]
        # Decode instruction data per PumpFun IDL

Instruction Discriminators

Most Solana programs use 8-byte discriminators (Anchor SHA256 hash of the instruction name). Match instruction.data[:8] against known values for each program.

# Common approach
PUMPFUN_CREATE = bytes.fromhex("181ec828051c0777")
PUMPFUN_BUY = bytes.fromhex("66063d1201daebea")
PUMPFUN_SELL = bytes.fromhex("33e685a4017f83ad")

disc = ix.data[:8]
if disc == PUMPFUN_BUY:
    # Parse buy parameters from remaining bytes
    ...

Warning: Discriminator values are program-specific and can change between program versions. Always verify against the current program IDL. See the pumpfun-mechanics skill for PumpFun-specific parsing.

Common Architecture: ShredStream + Yellowstone

Most production systems use both:

ShredStream (pre-execution)          Yellowstone gRPC (post-execution)
         │                                      │
         ▼                                      ▼
  Intent Detection                    Confirmation + Reconciliation
  "Wallet X is buying token Y"       "Buy succeeded, wallet now holds Z"
         │                                      │
         ▼                                      ▼
  Pre-compute Response                Execute / Update State
  (route, sign, prepare bundle)       (record PnL, update positions)

This gives you the speed advantage of ShredStream for signal detection while using Yellowstone for reliable state management.

Deployment Requirements

  • Public IP required — NAT breaks UDP shred delivery
  • Host networking — Docker bridge mode drops shred packets
  • UDP port 20000 open for incoming shreds
  • Co-locate near validators — Frankfurt, NY, Amsterdam, London, Tokyo recommended
  • Bare metal preferred — Cloud VMs add 1-5ms jitter from shared NICs

See references/deployment.md for full infrastructure guide.

Files

References

  • references/providers_compared.md — Jito ShredStream vs Shyft RabbitStream vs Triton Deshred
  • references/deployment.md — Infrastructure requirements, region selection, firewall configuration
  • references/proto_reference.md — ShredStream protobuf definitions and Entry parsing

Scripts

  • scripts/parse_shredstream_entries.py — Decode and analyze ShredStream gRPC entries
  • scripts/rabbitstream_monitor.py — Connect to Shyft RabbitStream for pre-execution transaction monitoring

How to use it

Copy the folder

Take agiprolabs/shredstream from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.

Install what it needs

The instructions reference docker. Without those the skill loads but fails at the first command.