Skip to main content

​
MCP Connection Issues

Symptoms: The tools are absent, or the host lists the server as failed.Solutions:
  1. Build first. The focused servers run from dist/, which does not exist until npm run build has succeeded.
  2. Use an absolute path. args must point at the full path to dist/index.js, not a relative one. The host does not run from your project directory.
  3. Restart the host. MCP hosts read their config at startup, so a running client will not pick up a new server.
  4. Check the JSON. A trailing comma or an unquoted key silently invalidates the whole config in most hosts.
Symptoms: The host reports the process died on launch.Cause: A required environment variable is missing. WDK_SEED is the only mandatory one, and the process refuses to start without it.Solutions:
  1. Confirm WDK_SEED is set in the env block of the host config, not in your shell (the host does not inherit your shell environment).
  2. Verify the mnemonic is a valid BIP-39 phrase with words separated by single spaces.
  3. Run the server by hand with the same env vars to see the error directly:
Symptoms: Streamable HTTP requests are rejected.Cause: MCP_AUTH_TOKEN is set on the server, so a bearer token is required on every request.Solutions:
  1. Send Authorization: Bearer <token> with the exact value of MCP_AUTH_TOKEN.
  2. If you did not intend to require auth, unset MCP_AUTH_TOKEN and restart. Only do that on a loopback interface.
Symptoms: spark_* tools work but wdk_* or kaleidoswap_* do not, or vice versa.Cause: You are connected to a focused server rather than the gateway. Each focused server exposes only its own domain.Solutions:
  1. Use kaleido-mcp to get every prefix over one connection.
  2. Or add each focused server to the host config separately.

​
Node and Wallet Errors

Symptoms: Wallet or channel tools fail while market data tools still work.Cause: No RGB Lightning Node is reachable at RLN_NODE_URL, which defaults to http://localhost:3001.Solutions:
  1. Start a node. The quickest route is the CLI:
  2. Verify it answers:
  3. If the node runs elsewhere, set RLN_NODE_URL to that address in the host config.
  4. See Node Environments for the Docker-backed setup.
Symptoms: The node responds, but wallet calls fail with an unlock or password error.Cause: An RGB Lightning Node must be unlocked after start before it will serve wallet operations.Solutions:
  1. Unlock through the CLI, or use the kaleido_node_unlock tool exposed by kaleido-mcp.
  2. Expect a wait: unlocking syncs the node before it becomes fully accessible.
Symptoms: KaleidoAgent in skill mode, or the kaleido_node_* tools, fail to execute.Cause: The binary is not on PATH where the process can see it.Solutions:
  1. Confirm it resolves: kaleido --version
  2. Set KALEIDO_BIN to the absolute path of the binary in the server or agent config.
  3. Install it per CLI Getting Started if it is missing.
Symptoms: Tools return empty balances for a wallet you know is funded.Solutions:
  1. Check the network. SPARK_NETWORK and the node’s own network must match where the funds are. A Regtest seed shows nothing on mainnet.
  2. Confirm the seed. A different mnemonic derives a different wallet, not an empty one.
  3. Wait for sync. A node that has just unlocked may not have caught up to the chain tip.
  4. Distinguish on-chain from Lightning. wdk_get_balances separates vanilla and colored UTXOs from the Lightning balance, and an asset in a channel is not an on-chain balance.

​
Swap Failures

Symptoms: An rfq_id is rejected when you try to use it.Cause: Quotes carry a short expiry, and an LLM turn plus a confirmation prompt can outlast it.Solutions:
  1. Re-quote immediately before initiating, rather than reusing an earlier rfq_id.
  2. Reduce the number of steps between quote and execution in your prompt or skill.
Symptoms: kaleidoswap_atomic_execute errors, or the status never leaves pending.Cause: An atomic swap needs both the DEX tools and the wallet tools, in order, plus channel liquidity on your side.Solutions:
  1. Confirm the sequence ran fully: init on the DEX, wdk_atomic_taker to whitelist the HTLC on your node, then execute. See the cross-server sequence.
  2. Check the asset is actually in a channel. On-chain balance alone cannot settle a Lightning swap.
  3. Poll kaleidoswap_atomic_status by payment_hash for the real state.
  4. If either side stalls, the timelocks expire and both sides recover their own funds. That is the designed outcome, not a lost swap.
Symptoms: A swap quotes or executes for far more or less than intended.Cause: Tools take raw amounts in the asset’s smallest unit, and assets have different precision. USDT commonly uses 6 decimals while BTC amounts are in satoshis.Solutions:
  1. Read kaleidoswap_get_assets for each asset’s precision before constructing an amount.
  2. Have the model state the raw amount and the display amount in its confirmation, so a mismatch is visible before you approve.

​
KaleidoAgent Issues

Symptoms: Loops run and report intended trades, but nothing executes.Cause: portfolio.dry_run is true. This is the default and is intentional.Solution: Set it to false only once the dry-run decisions have looked correct across several runs.
Symptoms: The agent was executing, then stopped submitting swaps.Cause: A risk limit tripped. Most often stop_loss_btc_sats, which halts all trading below its BTC threshold, or min_btc_reserve_sats.Solutions:
  1. Read GET /status for balances and recent runs.
  2. Compare them against the limits in agent.config.json.
  3. Fund the wallet or adjust the threshold deliberately, rather than removing the guard.
Symptoms: The portfolio is off target but no swap is proposed.Cause: Drift has not crossed rebalance_threshold_pct, which defaults to 5 percent.Solution: Lower the threshold, or trigger a run by hand with POST /run and {"task_id":"rebalance"}.
Symptoms: The agent starts but reasoning calls fail.Solutions:
  1. Confirm ANTHROPIC_API_KEY or OPENAI_API_KEY is set in .env.
  2. Confirm AGENT_PROVIDER matches the key you actually provided.

​
Rate Limits and Data

Symptoms: l402_get_price or l402_get_market_data fail intermittently.Cause: The free CoinGecko tier behind the public endpoints is rate limited.Solutions:
  1. Avoid calling price tools more than once every 30 seconds.
  2. Use l402_get_market_data to batch several assets into one call instead of looping.
  3. Set MPP_GATEWAY_URL to reach an authenticated MPP server if you need higher throughput.

​
Get Help

Check the FAQ for questions rather than errors. For anything else, report a problem through your preferred channel from the options below, including:
  1. Which surface and version (MCP server name, KaleidoAgent commit, or Desktop App version)
  2. The network you are on (Regtest, Signet, or mainnet)
  3. The tool name that failed and the error text
  4. Host and runtime versions (MCP client, Node.js)
  5. Your config with the seed and any keys removed

Telegram Community

Ask the community.

GitHub Issues

Report a bug on the relevant repository.

Email Support

Direct support for urgent issues.