Address forensic review-quorum findings and conform SVM paywall to upstream pattern:
- Drop dead solana:TESTNET FAUCET_URLS entry (Circle faucet does not
dispense USDC on Solana Testnet; Option B renders "No faucet configured."
for any unmapped chain).
- Revert all Solana Testnet recognition added by prior refactor pass
(SOLANA_NETWORK_REFS.TESTNET constant, getNetworkDisplayName branch,
isTestnetNetwork OR-branch). The paywall now recognizes one non-mainnet
SVM network — Devnet — matching upstream/main. No SVM dev expects
paywall Testnet support; upstream doesn't provide it.
- Unify "Need {tokenName} on {chainName}?" copy in SolanaPaywall and
AvmPaywall payment-required headers (were hardcoded "USDC"; matches
EvmPaywall).
Closes#2159
Add USDC.e (Bridged USDC) as the default stablecoin for HPP mainnet
(chain ID 190415) and HPP Sepolia (chain ID 181228) across all three SDKs:
- TypeScript: DEFAULT_STABLECOINS map in @x402/evm + regenerated NETWORK_DECIMALS in @x402/paywall
- Go: ChainIDHPP / ChainIDHPPSepolia constants, NetworkConfigs entries
- Python: NETWORK_CONFIGS dict in x402.mechanisms.evm
Contract details (verified on-chain):
- Token: USDC.e (Bridged USDC, FiatTokenV2_2)
- Address: 0x401eCb1D350407f13ba348573E5630B83638E30D (identical on both networks)
- EIP-712 domain: name="Bridged USDC", version="2"
- Decimals: 6
- EIP-3009: transferWithAuthorization supported
Chain details:
- HPP is an Arbitrum Orbit L2 operated by the HPP team
- Mainnet (190415): https://mainnet.hpp.io | Explorer: https://explorer.hpp.io
- Sepolia (181228): https://sepolia.hpp.io | Explorer: https://sepolia-explorer.hpp.io
- Portal: https://portal.hpp.io
USDC.e is the official bridged USDC issued by HPP, submitted by the HPP chain
team (chain-endorsed asset per DEFAULT_ASSETS.md).
The Permit2 erc20-approval client was hardcoding 1_000_000_000 / 100_000_000
as gas fee fallbacks. TypeScript already exposes these as
DEFAULT_MAX_FEE_PER_GAS / DEFAULT_MAX_PRIORITY_FEE_PER_GAS in evm/constants;
add the matching Python constants and use them. No behavior change.
Also adds the first unit tests for sign_erc20_approval_transaction covering
the fallback and successful estimate paths.
Tightens the viem floor in every package.json that lists viem (41 refs
across 40 files), refreshes all three lockfiles so every direct viem
resolution reads 2.48.11, and regenerates the paywall bundled templates
so the EVM/SVM/AVM bundles match the new viem chain database.
## Why floor-tightening matters
PR #2013 bumped lockfiles to 2.47.14, but later PRs (#2061, #2197, ...)
added sub-packages with ^2.39.x / ^2.43.x floors. pnpm re-resolved on
subsequent installs and the solver picked the lowest acceptable version
across constraints. ^-ranges with low floors do NOT preserve a bump --
lockfile refreshes erode the floor.
This PR fixes that incomplete tightening from #2013 by raising every
floor to ^2.48.11. The same regression will recur if any future PR adds
a viem-using sub-package with a lower floor. Gate criterion: new
packages must use ^2.48.11 (or tighter) for viem.
## Acceptance verification
- pnpm install --frozen-lockfile clean across typescript/,
examples/typescript/, and e2e/.
- pnpm format:check, lint:check, build, test green for typescript/.
- pnpm format:check, lint:check green for examples/typescript/.
- All direct viem resolutions read 2.48.11 in all three workspaces.
- Paywall bundled templates regenerated via
pnpm --filter @x402/paywall run build:paywall.
* feat: add TVM exact-payment mechanism to Python SDK
- add TVM exact client, server, and facilitator implementations
- add TVM signers, codecs, provider, and settlement helpers
- add unit and integration coverage for TON testnet/mainnet flows
- add Python examples and e2e wiring for TVM
* Enable TVM exact flows in Python e2e and tighten facilitator validation
Add TVM support to the Python e2e clients and servers, switch TVM usage to direct `ExactTvm*Scheme` registration after removing the old register helpers, and refresh docs/examples for testnet setup.
Also reject exact TVM settlements with excessively large attached TON amounts and cover the new validation with tests.
* fix: clarify minimum balance of the facilitator
* refactor(tvm): release settlement cache entry on confirmed settlement
On-chain seqno advances after a successful relay, so any retry of the same settlement BOC will fail at the W5 contract's seqno check regardless of the dedup reservation. Drop the entry as soon as the confirmation worker records success, so the invariant becomes "every is_duplicate=False is paired with a release" and TTL pruning is left as a crash safety-net only.
* feat(tvm): add TonAPI provider support
* feat(tvm): update e2e tests to make them compatible with new providers
* fix(tvm): update timeouts
* fix(tvm): increase http client timeout
* fix(tvm): align unit tests
* revert(e2e): remove .env-local config load
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(evm): add Radius Network and Radius Testnet default stablecoin (Permit2 + EIP-2612)
Add SBC as the default stablecoin for Radius Network (eip155:723487)
and Radius Testnet (eip155:72344) across Go, TypeScript, and Python SDKs.
SBC does not support EIP-3009, so Permit2 is the required transfer
method. SBC does support EIP-2612 permit() for gasless approval.
* docs: add Radius default-stablecoin entries to user-facing docs
* chore(paywall): update generated Radius decimals
The committed @x402/paywall bundle on main was last generated against an
older viem than the lockfile currently pins. pnpm install --frozen-lockfile
resolves viem 2.47.12 (via #2013), but the bundled src/evm/gen/template.ts
(and its Python + Go siblings) were built against an earlier resolution
that pre-dated chain definitions for Mezo, MegaETH, Stable, Radius (each
with a testnet variant), and 33 others. EvmPaywall.tsx throws
"Unsupported chain ID" at init for any eip155:<id> whose chain isn't in
the bundled viem/chains.
This PR:
1. Regenerates all nine auto-generated template files (TypeScript,
Python, and Go for EVM/SVM/AVM) against the current lockfile. No
source changes on any runtime path.
2. Teaches build:paywall to emit output matching the repo's formatters
(prettier / ruff), so generated files stay clean under pnpm run
format + uvx ruff format.
3. Adds a PR-time drift check
(.github/workflows/check_paywall_template.yml) so a future lockfile
or @x402/evm change that invalidates the bundle fails loudly
instead of shipping silently.
Scope is limited to @x402/paywall because it is the only package in this
repo that bakes viem chain definitions into a built artifact.
Verified on a fork-internal test PR running on ubuntu-latest
(Node 24, pnpm 10.7.0): check_format, check_lint, check_go, check_python,
check_package_lock, and the new check_paywall_template drift check all
pass. Additional local evidence:
- Chain-presence in regenerated bundle: Mezo 31612/31611, MegaETH
4326/6343, Stable 988/2201, Radius 723487/72344, Base/Arbitrum/
Polygon all present (mainnet/testnet ordering).
- Cross-platform determinism: a clean-slate rebuild on macOS produces
byte-identical output to ubuntu-latest CI.
- Tarball install into a throwaway consumer: all expected chain IDs
present in the published dist/esm/evm/index.js.
- Browser smoke test (Chrome, served from a local HTTP server):
EvmPaywall mounts without Unsupported-chain throw for
eip155:31611 (Mezo Testnet) and eip155:723487 (Radius); DOM
renders the expected token/chain labels.
Fixes#1971.
* fix(paywall): use dynamic token decimals instead of hardcoding 6
The EVM paywall displays incorrect amounts for any token that does not
use 6 decimal places. An 18-decimal token priced at 0.001 raw units
renders as $1,000,000,000 because the code divides by 10^6 everywhere.
Root cause: the paywall was built assuming USDC (6 decimals) is the
only payment token. This assumption appears in three places:
1. Server-side HTML generation (index.ts) divides the raw amount by
a hardcoded 10^6 to produce the display value injected into the
template.
2. Client-side balance display (EvmPaywall.tsx) calls formatUnits
with a hardcoded 6 and reads the token address from a 3-chain
USDC lookup table that returns 0 for any other chain.
3. The balance utility (utils.ts) only knows USDC addresses on
Ethereum, Base, and Base Sepolia — every other chain silently
shows a zero balance.
Fix:
Server-side (index.ts): look up the token's decimal precision from a
known-decimals map aligned with DEFAULT_STABLECOINS. Only non-6-decimal
chains need entries; everything else falls back to 6. This keeps the
server-side path simple with no RPC calls.
Client-side (EvmPaywall.tsx): read the token contract address from
the payment requirement's `asset` field (which the server already
populates) and query the standard ERC-20 decimals() function on-chain.
This works for any compliant token without maintaining a lookup table.
utils.ts: replace the USDC-specific getUSDCBalance (internal helper,
not part of the public API) with generic getTokenBalance and
getTokenDecimals functions that accept a token address parameter.
The function signature changes from 2 args to 3, but no external
consumer can import it — it is only used by the bundled React
component inside the paywall HTML template.
**Breaking change note:** the internal helper getUSDCBalance in utils.ts
is removed and replaced by getTokenBalance (different signature: takes
an explicit token address instead of looking it up by chain ID). This
function is not exported from the package's public API — it is only
consumed by the bundled EvmPaywall React component inside the paywall
HTML template. No server operator code calls it directly. However,
anyone who has forked the paywall and imports from the file path
(bypassing package exports) would need to update their import and
call site.
Regenerated Go, Python, and TS paywall templates.
**AI disclosure:** This PR was prepared with the assistance of a coding agent and
reviewed by Ryan R. Fox (an actual human) before submission.
* style: ruff format paywall templates
* fix(paywall): use DEFAULT_STABLECOINS lookup + BigInt formatUnits
Removes the parallel KNOWN_DECIMALS map in @x402/paywall/src/evm/index.ts
that mirrored DEFAULT_STABLECOINS from @x402/evm. The paywall now resolves
token decimals through the same registry that scheme implementations and
@x402/core's inline scheme dispatch already read from.
evmPaywall.generateHtml changes:
* getDefaultTokenDecimals helper looks up requirement.network directly in
@x402/evm's DEFAULT_STABLECOINS, with a 6 (USDC default) fallback when
the network is unknown.
* Atomic-to-display conversion uses Number(formatUnits(BigInt(amount),
decimals)) instead of parseFloat(amount) / 10**decimals. parseFloat
silently rounds sub-cent precision on real 18-decimal amounts before
the divide; BigInt + formatUnits preserves precision through the
conversion. BigInt also rejects non-integer atomic strings, which
matches the spec's atomic-integer contract.
@x402/paywall package.json: @x402/evm moves from devDependencies to
dependencies because the new server-side import of DEFAULT_STABLECOINS is
on the runtime path emitted into dist/.
@x402/evm now publicly re-exports DEFAULT_STABLECOINS from
./shared/defaultAssets so consumers can read the canonical default-asset
registry directly. Both @x402/evm and @x402/paywall get a minor bump
for the additive public API.
Adds 6 unit tests:
* getDefaultTokenDecimals returns 18 for Mezo Testnet (registry path)
* getDefaultTokenDecimals returns 6 for Base mainnet, asserted alongside
DEFAULT_STABLECOINS so an empty registry would fail the test
* getDefaultTokenDecimals falls back to 6 for unknown networks
* evmPaywall.generateHtml renders 1e15-atomic Mezo mUSD as
'amount: 0.001,' end-to-end (regression test for the /1e6 bug)
* evmPaywall.generateHtml renders 1e6-atomic Base USDC as 'amount: 1,'
* evmPaywall.generateHtml throws on a non-integer atomic ('1.5'),
pinning the BigInt strictness so a future revert to parseFloat fails
Closes#1979.
* refactor(paywall): generate evm decimals at build time, drop @x402/evm runtime dep
Addresses @phdargen's review: rather than promoting @x402/evm to a
runtime dependency of @x402/paywall, keep it a devDependency and
emit a codegen'd `src/evm/gen/decimals.ts` that mirrors just the
per-network `decimals` field of `DEFAULT_STABLECOINS`.
Pattern-consistent with the existing `src/evm/gen/template.ts` codegen
(HTML bundle) — same build step (`src/evm/build.ts`), same "AUTO-GENERATED
DO NOT EDIT" banner, same commit-the-generated-file workflow. The CI
drift check queued in #2054 (`check_paywall_template.yml`) will pick up
`decimals.ts` under the existing `typescript/packages/http/paywall/**`
trigger glob once that PR merges; in the meantime an in-process drift
invariant test in `network-handlers.test.ts` asserts `NETWORK_DECIMALS`
matches `DEFAULT_STABLECOINS` key-for-key.
Runtime behaviour is unchanged. `getDefaultTokenDecimals` now reads the
generated `NETWORK_DECIMALS` map instead of reading `DEFAULT_STABLECOINS`
at runtime; the same Mezo-18, Base-6, unknown→6 cases still hold, and
the existing 6 regression tests in `network-handlers.test.ts` pass
without modification. A seventh test pins the drift invariant described
above.
package.json: `@x402/evm` moved from `dependencies` to `devDependencies`.
Runtime module graph for `@x402/paywall` no longer pulls `@x402/evm`;
the only surviving `@x402/evm` references in paywall src are:
* `browserAdapter.ts` — `import type { ClientEvmSigner }` (type-only,
erased at build time)
* `EvmPaywall.tsx` — `ExactEvmScheme` (bundled into `gen/template.ts`
by esbuild, not part of the server-side runtime path)
* `build.ts` and `network-handlers.test.ts` — build-time / test-time
consumers; devDep is available in both.
Both the root `typescript/pnpm-lock.yaml` and `examples/typescript/pnpm-lock.yaml`
are updated to reflect the dep move — both workspaces reference the
paywall package manifest via `workspace:~`.
* add mcp discovery support
* ci fixes and changelog
* rename changelog
* fix failing test
* fail in TS on transport mismatch, return extension info on settlement failure in python
* add e2e
* fix mcp payment payload parsing
* fix python parsing
* add transport to examples and docs
* docs: add Documentation Checklist + surface chain-contribution flow
DEFAULT_ASSETS.md's 'Adding a New Chain' numbered flow stopped at 'Update
all three SDKs' → 'Submit a PR' and did not mention the user-facing docs
surfaces that also enumerate every chain with default-asset support. As a
result, recent onboarding PRs have shipped with stale docs (e.g. the
Radius PR #2038 updated the SDKs but not the docs table or Go README).
The existing Cross-SDK Checklist at the bottom enumerates only the three
code registries, mirroring the same gap.
network-and-token-support.mdx has a parallel discoverability problem:
the pointer to DEFAULT_ASSETS.md used a repo-relative path that resolves
to a broken link on docs.x402.org, and the pointer sat inside the
'Default Assets for Dollar-String Pricing' intro rather than under
'Adding Support for New Networks' where a would-be contributor actually
looks. The 'Adding' section itself described only runtime registration
(server.register from consumer code) and did not mention the separate
upstream-contribution path, and it also contained a 'Running Your Own
Facilitator' subsection that is about facilitator infrastructure, not
about adding networks — a topical mismatch. The page also ordered the
'Facilitators' section before 'Adding Support for New Networks', which
inverts the natural read → use → extend → operate narrative.
Changes:
* DEFAULT_ASSETS.md:
- Promote 'Update user-facing documentation' to a numbered step in
'Adding a New Chain' (new step 3, between 'Update all three SDKs'
and 'Submit a PR'), with a cross-reference to the new checklist.
- Append a new 'Documentation Checklist' section below the existing
Cross-SDK Checklist, naming the two hot docs surfaces
(network-and-token-support.mdx and go/mechanisms/evm/README.md)
that must also be updated on a new chain onboarding.
* docs/core-concepts/network-and-token-support.mdx:
- Remove the oddly-placed 'See DEFAULT_ASSETS.md' link from the
Dollar-String Pricing intro; replace with an in-page anchor
pointing at the new contribution subsection.
- Split 'Adding Support for New Networks' into two subsections:
'Runtime Registration' (the existing content, unchanged) and
a new 'Contributing a New Default Asset' subsection that links
to DEFAULT_ASSETS.md via an absolute github.com URL (so the
link works on the published docs site) and explicitly names
the cross-SDK + docs checklist.
- Move 'Running Your Own Facilitator' out of 'Adding Support for
New Networks' into the existing 'Facilitators' section, where
it belongs topically.
- Swap top-level section order: 'Adding Support for New Networks'
now precedes 'Facilitators', matching the read → use → extend →
operate narrative.
No code changes. Documentation only.
* docs: revert DEFAULT_ASSETS.md, unify SDK READMEs to link docs
Address review feedback: the Mintlify agent (.github/workflows/update-docs.yml)
keeps docs/ in sync with code, so contributors should only update the three SDK
constants files. Remove the Documentation Checklist and Step 3 from
DEFAULT_ASSETS.md. Apply a consistent two-link block to all three SDK READMEs
(Go/TS/Python) pointing at the mdx default-assets list and the mdx
Adding Support for New Networks section. Soften the mdx Contributing a New
Default Asset blurb to drop manual mdx/README update instructions.
Implement the spec change from PR #1682, adding optional seller-defined
memo support to the SVM exact scheme across all three SDKs.
Client changes (TypeScript, Go, Python - both v1 and v2):
- When extra.memo is present, use its UTF-8 value as memo instruction data
- When absent, fall back to existing random 16-byte nonce behavior
- Validate memo does not exceed MAX_MEMO_BYTES (256) limit
Facilitator changes (TypeScript, Go, Python - both v1 and v2):
- When extra.memo is present in requirements, verify exactly one memo
instruction exists and its data matches the expected value
- Return specific error codes: memo_count / memo_mismatch
Constants added:
- MAX_MEMO_BYTES / MaxMemoBytes (256) across all languages
- ERR_MEMO_MISMATCH, ERR_MEMO_COUNT (Python/Go)
- ErrMemoExceedsMaxSize (Go client)
Tests added:
- TypeScript: 4 memo tests in duplicateTx.test.ts
- Go: TestSellerMemo with 4 subtests in duplicate_tx_test.go
- Python: TestSellerMemo class with 4 tests in test_duplicate_tx.py
Closes#1682
The EVM paywall hardcodes "USDC" as the token label in all display
text. The payment requirements already carry the token name in
extra.name (populated by the server from the default asset registry),
but the paywall ignores it.
Read extra.name from the first payment requirement and use it for
all user-facing token text. Falls back to "USDC" when extra.name
is absent (backward compatible with EIP-3009 tokens).
5 display strings updated:
- Header: "please pay $X {tokenName}" (was "USDC")
- Testnet prompt: "Need {tokenName} on {chainName}?" (was "USDC")
- Balance: "$X {tokenName}" (was "USDC")
- Amount row: "$X {tokenName}" (was "USDC")
- Error: "Make sure you have {tokenName} on {chainName}" (was "USDC")
Regenerated Go, Python, and TS paywall templates.
* test(python): add missing sync test scenarios to existing test_server.py
Add edge case coverage for: missing/malformed _meta, no matching
requirements, MCPToolResult direct return, non-dict handler return,
structuredContent preservation, empty accepts validation, verification
failure without reason, and hook context field verification.
* fix: apply ruff format to test_server.py
* fix formatting: match existing test style, inline data construction, fix import order
Add native USDC (0x3c499c542cEF5E3811e1192ce70d8cC03d5c3359) as the
default stablecoin for Polygon mainnet across all three SDKs:
- Go: ChainIDPolygon constant, v2 NetworkConfigs, v1 NetworkConfigs
- TypeScript: v2 stablecoins map in ExactEvmScheme
- Python: v2 NETWORK_CONFIGS with supported_assets
Update Go v1 unit tests that previously used "polygon" as a network
without a default asset config. Replace with "iotex" and add positive
test coverage for Polygon config and asset resolution.
Contract details (verified on-chain via PolygonScan):
- Token: Circle native USDC (FiatTokenV2_2)
- EIP-712 domain: name="USD Coin", version="2"
- EIP-3009: transferWithAuthorization supported
- x402 proxy deployed at canonical CREATE2 address
* Fix race condition in lazy facilitator initialization for FastAPI and Flask middleware
Under concurrent requests, multiple coroutines/threads could see init_done=False
simultaneously and all call http_server.initialize(), causing redundant init calls
and inconsistent error propagation.
- FastAPI: guard with asyncio.Lock using double-checked locking pattern
- Flask: guard with threading.Lock using double-checked locking pattern
- Add concurrency tests proving single-flight initialization for both frameworks
- Add tests verifying failed init does not permanently block retries
Closes#1584
* fix: remove unused variables to pass ruff lint
* fix(go): add omitempty to QueryInput/BodyInput.Method to prevent spurious enum validation failure
When external JSON omits info.input.method (valid when method is not in the schema's
required list), Go zero-initializes the Method field to "" and re-marshaling produces
"method":"", which fails the enum constraint in ValidateDiscoveryExtension.
Adding omitempty ensures absent method fields remain absent after round-trip serialization.
Fixes#1830
Made-with: Cursor
* enforce method is required in TS and Python discovery extension schemas
Per spec, the method field must always be present in a discovery extension's
info.input. Add "method" to the required array in schemas built by the TS
and Python resource service builders, so ValidateDiscoveryExtension rejects
extensions missing method with a clear error.
Also reverts the Go omitempty workaround from the prior commit; the Go SDK
was already producing a validation error for absent method (via the enum
constraint), and the schema-level required enforcement is the correct fix.
Closes#1830
Made-with: Cursor
* style: format resourceService.ts type union per Prettier
Made-with: Cursor
* fix: auto-populate HTTP method from adapter in all SDKs
The SDK entry points (processHTTPRequest, requiresPayment,
processSettlement) now derive the HTTP method from the adapter
when it is not explicitly set on the request context. This ensures
direct x402HTTPResourceServer users do not need to manually assign
the method field — the SDK computes it at runtime, matching the
behavior middleware users already get.
Made-with: Cursor
* test: add declare → enrich → validate flow tests for bazaar extension
Verify the real middleware path: extensions are declared without method,
enrichDeclaration injects method from the HTTP context at request time,
and the enriched extension passes schema validation. Covers GET and
POST in all three SDKs (TypeScript, Go, Python).
Made-with: Cursor
* feat(evm): add Stable mainnet (chain ID 988) network support
* chore: add changelog fragments for Stable mainnet support
* feat(evm): add Stable mainnet to Go v1 legacy network config
* fix: add duplicate settlement mitigation for SVM across all SDKs
Introduce an in-memory SettlementCache to prevent a race condition on
Solana where the same payment transaction could be settled multiple
times before on-chain confirmation. The cache is shared across V1 and
V2 facilitator schemes so that cross-version duplicates are caught.
- Add SettlementCache with 120s TTL in TypeScript, Python, and Go
- Integrate cache check into facilitator settle paths (V1 and V2)
- Add unit tests for duplicate detection in all three languages
- Share cache across V1/V2 in Go facilitator example
- Add duplicate settlement spec to scheme_exact_svm.md
- Document SettlementCache in SVM READMEs for all three SDKs
- Document the race condition in go/FACILITATOR.md Security section
- Add merchant-facing guidance in docs/core-concepts/client-server.md
and docs/core-concepts/facilitator.md for self-settling servers
- Fix JSDoc lint errors in TypeScript settlement-cache.ts
* refactor: optimize SettlementCache prune with early break on ordered entries
Leverage insertion-order guarantees of Map (TS) and dict (Python) to
break early once a non-expired entry is encountered, avoiding a full
scan of the cache on every isDuplicate call. Go is left unchanged since
map iteration order is not guaranteed.
Also adds prune-specific unit tests for all three SDKs covering:
- mixed expired/fresh entries
- all expired
- none expired
- insertion-order early-break correctness (Python)
* docs: add changelog entries for SVM duplicate settlement fix