Skip to content

ggoldani/stellar-mcp

BranchesTags

Repository files navigation

StellarMCP

MCP server for Stellar: accounts, payments, XDR, Horizon/Soroban RPC, AMM liquidity, SEP anchors, and Soroban (simulate, invoke, events, deploy, contract state). Intended for agents and IDE integrations (Cursor, Claude Desktop, etc.) with strict validation, normalized errors, and redacted _debug fields.

All tools (names, descriptions, parameters): docs/TOOLS.md — auto-generated from the same tools/list your MCP client sees; attach or paste it when asking an agent how to use this server.

Requirements: Node.js ≥ 20. · Source: github.com/ggoldani/stellar-mcp

Table of contents

Quick start

From a clone (this repository)

git clone https://github.com/ggoldani/stellar-mcp.git
cd stellar-mcp
npm install
npm run build

Run the server (see Run the server), then connect your MCP client.

From npm

Package on npm: @ggoldani/stellarmcp (scoped package).

The published tarball includes build/src, templates/, generator inputs (src/lib/errors.ts, redact.ts), .env.example, and docs — enough to run the server and stellarmcp-generate without cloning.

npm install @ggoldani/stellarmcp
npx stellarmcp
# or: npx @ggoldani/stellarmcp
  • Env template: node_modules/@ggoldani/stellarmcp/.env.example → copy to your project as .env if needed.
  • MCP stdio when the package is a dependency (example): "args": ["${workspaceFolder}/node_modules/@ggoldani/stellarmcp/build/src/index.js"].

Prefer GitHub if you are developing or patching this repository (clone).

Sanity check (first call)

Use this to confirm wiring before deeper integration:

  1. HTTP mode: with MCP_TRANSPORT=http-sse running, open GET /health (expect JSON with network, horizonReachable, rpcReachable).
  2. Any MCP host: call tool stellar_get_fee_stats with {}. Expect fields like baseFee, p99, recommendedFee (see Example tool calls).

If that works, Horizon is reachable and the server is usable for read-only tools without a secret key.

For every tool name and argument shape, use docs/TOOLS.md (or your host’s MCP tool picker).

Run the server

Mode Command Notes
stdio (default) npm run start:stdio or node build/src/index.js Best for Claude Desktop, Cursor stdio, local agents.
Streamable HTTP MCP_TRANSPORT=http-sse PORT=3000 npm run start:http MCP endpoint: http://<host>:<PORT>/mcp. Health: GET /health (JSON: network, Horizon/RPC reachability, version).

After npm run build, the entrypoint is build/src/index.js.

Connect your MCP client

stdio (recommended locally)

Point your host at Node and the built entrypoint. Example shape (paths must be absolute or use your editor’s variable such as ${workspaceFolder}):

{
  "mcpServers": {
    "stellarmcp": {
      "type": "stdio",
      "command": "node",
      "args": ["/ABSOLUTE/PATH/TO/stellarmcp/build/src/index.js"],
      "env": {
        "MCP_TRANSPORT": "stdio",
        "STELLAR_NETWORK": "testnet"
      }
    }
  }
}
  • Cursor: project config ~/.cursor/mcp.json or .cursor/mcp.json — see Cursor MCP docs (interpolation: ${workspaceFolder}, ${env:VAR}).
  • Claude Desktop: claude_desktop_config.json under the same mcpServers pattern.
  • Optional: set "envFile": "${workspaceFolder}/.env" (stdio only in Cursor) to load secrets from a local .env without committing it.

Never commit real STELLAR_SECRET_KEY or anchor tokens.

HTTP / SSE

  1. Start the server with MCP_TRANSPORT=http-sse and choose PORT.
  2. In the client, register the remote server, for example:
{
  "mcpServers": {
    "stellarmcp": {
      "url": "http://localhost:3000/mcp"
    }
  }
}
  1. Verify GET http://localhost:3000/health before debugging MCP calls.

Environment variables

Authoritative template: copy [.env.example](./.env.example) to .env and edit.

Variable Required Purpose
MCP_TRANSPORT No stdio (default) or http-sse.
STELLAR_NETWORK No testnet (default) or mainnet.
STELLAR_HORIZON_URL No Horizon base URL (https; host must be default-allowed or listed in STELLAR_ALLOWED_HOSTS).
STELLAR_RPC_URL No Soroban RPC URL (same rules as Horizon).
STELLAR_SEP38_URL No Anchor SEP-38 endpoint for quotes.
STELLAR_ALLOWED_HOSTS No Comma-separated extra hostnames allowed for the three URL vars above.
STELLAR_TRUSTED_ANCHOR_DOMAINS No Comma-separated domains allowed for anchor/TOML and SEP flows.
STELLAR_SECRET_KEY No Signing key; omit for read-only / unsigned-XDR-only operation.
STELLAR_AUTO_SIGN_POLICY No safe, guarded, or expert — see Security.
STELLAR_AUTO_SIGN No Legacy if policy unset.
STELLAR_AUTO_SIGN_LIMIT No Legacy / used with guarded (must be > 0).
STELLAR_USDC_ISSUER No USDC issuer for valuation in guarded signing (default is network USDC).
PORT No HTTP server port (default 3000).
STELLAR_REQUEST_TIMEOUT_MS No Upstream request timeout (max 30000).
MCP_HTTP_* No Rate limit, concurrency, max POST body, TRUST_PROXY — see .env.example.
STELLAR_META_* No Disk cache + XDR size limits for meta tools — see .env.example.

Custom Horizon/RPC/SEP-38 URLs must use https and non-private hosts; unknown hosts require STELLAR_ALLOWED_HOSTS.

Security

  1. Secrets: Treat STELLAR_SECRET_KEY like production key material. Prefer env or envFile; never commit secrets; restrict file permissions on .env.
  2. Auto-sign policy (write tools: payments, trustlines, liquidity, set_options, fee bump, Soroban invoke/deploy, relevant SEP flows):
    • safe (recommended): unsigned mode — write tools return unsigned XDR for external signing when applicable.
    • guarded: auto-sign enabled only with STELLAR_AUTO_SIGN_LIMIT > 0 and USDC-based valuation rules; fails closed when value cannot be bounded.
    • expert: unlimited auto-sign — use only with full awareness of risk.
  3. Network exposure: If you expose HTTP/SSE, use TLS in front, sane firewall rules, and review MCP_HTTP_*. MCP_HTTP_TRUST_PROXY=true only behind a trusted proxy that sets X-Forwarded-For correctly.
  4. Allowlists: Use STELLAR_ALLOWED_HOSTS and STELLAR_TRUSTED_ANCHOR_DOMAINS when pointing at non-default infrastructure or anchors.
  5. Read-only deployments: Omit STELLAR_SECRET_KEY; read and simulate tools still work; writes yield unsigned payloads or explicit errors.

Tools reference

Full detail for agents (every tool, descriptions, parameter tables from live tools/list): docs/TOOLS.md — auto-generated; refresh with npm run docs:tools after you change tools.

Summary below: Read = no transaction submission by this server; Write = may build/submit transactions, call Friendbot, or initiate anchor flows.

Accounts & history

Tool Type Description
stellar_get_account Read Balances, signers, flags, subentries, minimum balance.
stellar_get_account_history Read Paginated transaction history (limit, optional cursor).
stellar_fund_account Write Testnet only — funds via Friendbot HTTP (10k test XLM).
stellar_set_options Write Account options (signers, weights, flags); unsigned unless policy allows signing.

Payments & fees

Tool Type Description
stellar_submit_payment Write Payment; hash or unsigned XDR per policy.
stellar_submit_fee_bump_transaction Write Fee-bump an existing transaction (sponsor pays fee).

Assets & AMM

Tool Type Description
stellar_create_trustline Write Create trustline for non-native asset.
stellar_deposit_liquidity Write Deposit into classic AMM pool.
stellar_withdraw_liquidity Write Withdraw from classic AMM pool.

Network

Tool Type Description
stellar_get_fee_stats Read Fee stats + recommended fee (stroops).

XDR

Tool Type Description
stellar_xdr_types Read List XDR type names (optional prefix).
stellar_xdr_json_schema Read Draft-7 JSON Schema for a type.
stellar_xdr_guess Read Candidate types for a base64 XDR blob.
stellar_xdr_encode Read JSON → base64 XDR for a named type.
stellar_decode_xdr Read Classic transaction XDR → JSON (uses configured network passphrase).

Historical meta (Horizon-first, Soroban RPC fallback, bounded XDR)

Tool Type Description
stellar_get_ledger_meta Read Closed ledger header/metadata XDR (+ optional cache metadata).
stellar_get_transaction_meta Read Tx envelope/result/result-meta/fee-meta; optional operationIndex.

SEP / anchors

Tool Type Description
stellar_get_anchor_toml Read Fetch/parse stellar.toml for SEP discovery.
stellar_sep10_auth Write SEP-10 challenge → JWT (requires signer / secret where applicable).
stellar_sep6_transfer Write SEP-6 deposit/withdraw initiation.
stellar_sep12_customer Mixed GET/PUT KYC against anchor SEP-12 server.
stellar_sep24_interactive Write SEP-24 interactive URL for deposit/withdraw.
stellar_sep31_remittance Write SEP-31 remittance initiation.
stellar_get_sep38_quote Read SEP-38 indicative quote / rate metadata.

Soroban

Tool Type Description
stellar_soroban_simulate Read Simulate contract call (footprint, fees, events); does not submit.
stellar_soroban_invoke Write Simulate, assemble, sign/submit per policy.
stellar_soroban_get_events Read Contract events from RPC (startLedger, filters, limit).
stellar_soroban_deploy Write Deploy WASM from disk path; submit per policy.
stellar_soroban_read_state Read Direct getLedgerEntries for a contract data key.

Example tool calls

JSON shapes are illustrative; your MCP host sends tools/call with name + arguments.

Fee stats (testnet)

{ "name": "stellar_get_fee_stats", "arguments": {} }

XDR: list types → schema → encode

{ "name": "stellar_xdr_types", "arguments": { "prefix": "Transaction" } }
{ "name": "stellar_xdr_json_schema", "arguments": { "type": "TransactionEnvelope" } }
{ "name": "stellar_xdr_encode", "arguments": { "type": "TransactionEnvelope", "json": "<JSON string>" } }
{ "name": "stellar_xdr_guess", "arguments": { "xdr": "<base64>" } }
{ "name": "stellar_decode_xdr", "arguments": { "xdr": "<base64 classic transaction XDR>" } }

Historical meta

{ "name": "stellar_get_ledger_meta", "arguments": { "ledgerSequence": 123456, "maxXdrCharsPerField": 8192 } }
{
  "name": "stellar_get_transaction_meta",
  "arguments": {
    "transactionHash": "<64-char hex>",
    "operationIndex": 0,
    "maxXdrCharsPerField": 8192
  }
}

For full argument schemas, use your client’s tool list / schema UI or inspect Zod definitions under src/tools/.

Soroban contract MCP generator

Generate a standalone Node MCP package (stdio) from a deployed contract’s WASM (contractspecv0) or from a spec JSON manifest.

Inputs

  1. WASM — Spec.fromWasm reads the custom section (same idea as @stellar/stellar-sdk contract specs).
  2. Spec JSON — format stellarmcp-contract-spec-v1 with entries[] of base64 ScSpecEntry XDR values.

CLI (from this repo after npm run build)

node build/src/generator/cli.js --input path/to/contract.wasm --out ./my-contract-mcp --name my-contract-mcp --alias mytoken
# or
node build/src/generator/cli.js --input path/to/spec.json --out ./my-contract-mcp --name my-contract-mcp --alias mytoken

Published installs: stellarmcp-generate (same flags).

Then in the output directory:

npm install && npm run build
STELLAR_CONTRACT_ID=C... STELLAR_NETWORK=testnet node build/src/index.js

Non-goals (current generator)

  • No generated HTTP/SSE transport in the scaffold (stdio only).
  • Exotic spec shapes may emit loose Zod at the edges; simulation on RPC remains authoritative.
  • Multi-contract workspaces / auto-publish are out of scope for the generator CLI.

Versioning artifacts in generated code: GENERATOR_ARTIFACT_VERSION, SPEC_FINGERPRINT, etc. — regenerate when upgrading stellarmcp or changing the contract interface.

Troubleshooting

Issue What to check
Client cannot start server node -v ≥ 20; npm run build; path to build/src/index.js is correct.
Tools error on Horizon/RPC STELLAR_NETWORK matches intended network; URLs https; hosts allowlisted if custom.
Signing / “unsigned only” STELLAR_AUTO_SIGN_POLICY=safe returns unsigned XDR; add key + policy only if you intend auto-sign.
HTTP mode GET /health; MCP at POST/GET /mcp per Streamable HTTP transport; see MCP Logs in Cursor.
Meta cache Writable STELLAR_META_CACHE_DIR or temp dir; freshness.cacheWriteOk may be false on read-only FS.

Development & testing

npm run typecheck
npm run test             # build + unit tests + docs/TOOLS.md drift check
npm run docs:tools       # regenerate docs/TOOLS.md after changing MCP tools
npm run verify:phase:c   # full maintainer gate (see package.json)

Useful smokes:

  • npm run smoke:phase1 — config + stdio + HTTP wiring.
  • npm run smoke:testnet:readonly — real testnet reads + Soroban via MCP stdio (no secret key).
  • npm run smoke:tier1:friendbot — Friendbot-funded flows on testnet.

Inspired by stellarskills

Knowledge layer: stellarskills. Execution layer: StellarMCP.

License: MIT (see repository).

About

No description, website, or topics provided.

Resources

Readme

License

MIT license
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages