@ic402/mcp 2.5.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/LICENSE ADDED
@@ -0,0 +1,190 @@
1
+ Apache License
2
+ Version 2.0, January 2004
3
+ http://www.apache.org/licenses/
4
+
5
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
6
+
7
+ 1. Definitions.
8
+
9
+ "License" shall mean the terms and conditions for use, reproduction,
10
+ and distribution as defined by Sections 1 through 9 of this document.
11
+
12
+ "Licensor" shall mean the copyright owner or entity authorized by
13
+ the copyright owner that is granting the License.
14
+
15
+ "Legal Entity" shall mean the union of the acting entity and all
16
+ other entities that control, are controlled by, or are under common
17
+ control with that entity. For the purposes of this definition,
18
+ "control" means (i) the power, direct or indirect, to cause the
19
+ direction or management of such entity, whether by contract or
20
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
21
+ outstanding shares, or (iii) beneficial ownership of such entity.
22
+
23
+ "You" (or "Your") shall mean an individual or Legal Entity
24
+ exercising permissions granted by this License.
25
+
26
+ "Source" form shall mean the preferred form for making modifications,
27
+ including but not limited to software source code, documentation
28
+ source, and configuration files.
29
+
30
+ "Object" form shall mean any form resulting from mechanical
31
+ transformation or translation of a Source form, including but
32
+ not limited to compiled object code, generated documentation,
33
+ and conversions to other media types.
34
+
35
+ "Work" shall mean the work of authorship, whether in Source or
36
+ Object form, made available under the License, as indicated by a
37
+ copyright notice that is included in or attached to the work
38
+ (an example is provided in the Appendix below).
39
+
40
+ "Derivative Works" shall mean any work, whether in Source or Object
41
+ form, that is based on (or derived from) the Work and for which the
42
+ editorial revisions, annotations, elaborations, or other modifications
43
+ represent, as a whole, an original work of authorship. For the purposes
44
+ of this License, Derivative Works shall not include works that remain
45
+ separable from, or merely link (or bind by name) to the interfaces of,
46
+ the Work and Derivative Works thereof.
47
+
48
+ "Contribution" shall mean any work of authorship, including
49
+ the original version of the Work and any modifications or additions
50
+ to that Work or Derivative Works thereof, that is intentionally
51
+ submitted to the Licensor for inclusion in the Work by the copyright owner
52
+ or by an individual or Legal Entity authorized to submit on behalf of
53
+ the copyright owner. For the purposes of this definition, "submitted"
54
+ means any form of electronic, verbal, or written communication sent
55
+ to the Licensor or its representatives, including but not limited to
56
+ communication on electronic mailing lists, source code control systems,
57
+ and issue tracking systems that are managed by, or on behalf of, the
58
+ Licensor for the purpose of discussing and improving the Work, but
59
+ excluding communication that is conspicuously marked or otherwise
60
+ designated in writing by the copyright owner as "Not a Contribution."
61
+
62
+ "Contributor" shall mean Licensor and any individual or Legal Entity
63
+ on behalf of whom a Contribution has been received by the Licensor and
64
+ subsequently incorporated within the Work.
65
+
66
+ 2. Grant of Copyright License. Subject to the terms and conditions of
67
+ this License, each Contributor hereby grants to You a perpetual,
68
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
69
+ copyright license to reproduce, prepare Derivative Works of,
70
+ publicly display, publicly perform, sublicense, and distribute the
71
+ Work and such Derivative Works in Source or Object form.
72
+
73
+ 3. Grant of Patent License. Subject to the terms and conditions of
74
+ this License, each Contributor hereby grants to You a perpetual,
75
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
76
+ (except as stated in this section) patent license to make, have made,
77
+ use, offer to sell, sell, import, and otherwise transfer the Work,
78
+ where such license applies only to those patent claims licensable
79
+ by such Contributor that are necessarily infringed by their
80
+ Contribution(s) alone or by combination of their Contribution(s)
81
+ with the Work to which such Contribution(s) was submitted. If You
82
+ institute patent litigation against any entity (including a
83
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
84
+ or a Contribution incorporated within the Work constitutes direct
85
+ or contributory patent infringement, then any patent licenses
86
+ granted to You under this License for that Work shall terminate
87
+ as of the date such litigation is filed.
88
+
89
+ 4. Redistribution. You may reproduce and distribute copies of the
90
+ Work or Derivative Works thereof in any medium, with or without
91
+ modifications, and in Source or Object form, provided that You
92
+ meet the following conditions:
93
+
94
+ (a) You must give any other recipients of the Work or
95
+ Derivative Works a copy of this License; and
96
+
97
+ (b) You must cause any modified files to carry prominent notices
98
+ stating that You changed the files; and
99
+
100
+ (c) You must retain, in the Source form of any Derivative Works
101
+ that You distribute, all copyright, patent, trademark, and
102
+ attribution notices from the Source form of the Work,
103
+ excluding those notices that do not pertain to any part of
104
+ the Derivative Works; and
105
+
106
+ (d) If the Work includes a "NOTICE" text file as part of its
107
+ distribution, then any Derivative Works that You distribute must
108
+ include a readable copy of the attribution notices contained
109
+ within such NOTICE file, excluding any notices that do not
110
+ pertain to any part of the Derivative Works, in at least one
111
+ of the following places: within a NOTICE text file distributed
112
+ as part of the Derivative Works; within the Source form or
113
+ documentation, if provided along with the Derivative Works; or,
114
+ within a display generated by the Derivative Works, if and
115
+ wherever such third-party notices normally appear. The contents
116
+ of the NOTICE file are for informational purposes only and
117
+ do not modify the License. You may add Your own attribution
118
+ notices within Derivative Works that You distribute, alongside
119
+ or as an addendum to the NOTICE text from the Work, provided
120
+ that such additional attribution notices cannot be construed
121
+ as modifying the License.
122
+
123
+ You may add Your own copyright statement to Your modifications and
124
+ may provide additional or different license terms and conditions
125
+ for use, reproduction, or distribution of Your modifications, or
126
+ for any such Derivative Works as a whole, provided Your use,
127
+ reproduction, and distribution of the Work otherwise complies with
128
+ the conditions stated in this License.
129
+
130
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
131
+ any Contribution intentionally submitted for inclusion in the Work
132
+ by You to the Licensor shall be under the terms and conditions of
133
+ this License, without any additional terms or conditions.
134
+ Notwithstanding the above, nothing herein shall supersede or modify
135
+ the terms of any separate license agreement you may have executed
136
+ with Licensor regarding such Contributions.
137
+
138
+ 6. Trademarks. This License does not grant permission to use the trade
139
+ names, trademarks, service marks, or product names of the Licensor,
140
+ except as required for reasonable and customary use in describing the
141
+ origin of the Work and reproducing the content of the NOTICE file.
142
+
143
+ 7. Disclaimer of Warranty. Unless required by applicable law or
144
+ agreed to in writing, Licensor provides the Work (and each
145
+ Contributor provides its Contributions) on an "AS IS" BASIS,
146
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147
+ implied, including, without limitation, any warranties or conditions
148
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149
+ PARTICULAR PURPOSE. You are solely responsible for determining the
150
+ appropriateness of using or redistributing the Work and assume any
151
+ risks associated with Your exercise of permissions under this License.
152
+
153
+ 8. Limitation of Liability. In no event and under no legal theory,
154
+ whether in tort (including negligence), contract, or otherwise,
155
+ unless required by applicable law (such as deliberate and grossly
156
+ negligent acts) or agreed to in writing, shall any Contributor be
157
+ liable to You for damages, including any direct, indirect, special,
158
+ incidental, or consequential damages of any character arising as a
159
+ result of this License or out of the use or inability to use the
160
+ Work (including but not limited to damages for loss of goodwill,
161
+ work stoppage, computer failure or malfunction, or any and all
162
+ other commercial damages or losses), even if such Contributor
163
+ has been advised of the possibility of such damages.
164
+
165
+ 9. Accepting Warranty or Additional Liability. While redistributing
166
+ the Work or Derivative Works thereof, You may choose to offer,
167
+ and charge a fee for, acceptance of support, warranty, indemnity,
168
+ or other liability obligations and/or rights consistent with this
169
+ License. However, in accepting such obligations, You may act only
170
+ on Your own behalf and on Your sole responsibility, not on behalf
171
+ of any other Contributor, and only if You agree to indemnify,
172
+ defend, and hold each Contributor harmless for any liability
173
+ incurred by, or claims asserted against, such Contributor by reason
174
+ of your accepting any such warranty or additional liability.
175
+
176
+ END OF TERMS AND CONDITIONS
177
+
178
+ Copyright 2026 ic402 contributors
179
+
180
+ Licensed under the Apache License, Version 2.0 (the "License");
181
+ you may not use this file except in compliance with the License.
182
+ You may obtain a copy of the License at
183
+
184
+ http://www.apache.org/licenses/LICENSE-2.0
185
+
186
+ Unless required by applicable law or agreed to in writing, software
187
+ distributed under the License is distributed on an "AS IS" BASIS,
188
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
189
+ See the License for the specific language governing permissions and
190
+ limitations under the License.
package/README.md ADDED
@@ -0,0 +1,326 @@
1
+ # @ic402/mcp
2
+
3
+ An [MCP](https://modelcontextprotocol.io) (Model Context Protocol) server that exposes an
4
+ ic402-enabled ICP canister to an AI agent over **stdio**. It lets a model probe x402-gated
5
+ endpoints, open and settle streaming micropayment sessions, fetch encrypted content, drive the
6
+ ERC-8004 / marketplace flows, and — when an operator explicitly opts in — perform
7
+ controller-gated admin and signing actions.
8
+
9
+ The server identifies itself to MCP clients as:
10
+
11
+ ```
12
+ name: ic402
13
+ version: 2.5.2
14
+ ```
15
+
16
+ (see `src/index.ts`, the `new McpServer({ name: 'ic402', version: '2.5.2' })` registration).
17
+
18
+ > **Security note up front.** This server is driven by an LLM whose inputs may be influenced by
19
+ > untrusted web content (prompt injection) while it holds a controller identity capable of
20
+ > signing value transfers from the canister's own EVM address. The security posture is therefore
21
+ > **operator-set, out-of-band** — see [Security model](#security-model). The LLM cannot loosen it.
22
+
23
+ ## Build & run
24
+
25
+ The package builds with plain `tsc` (output goes to `dist/`, with `dist/index.js` as the
26
+ `ic402-mcp` bin):
27
+
28
+ ```bash
29
+ # From the repo root — builds the client SDK, this MCP server, and the demo client:
30
+ pnpm build:demo
31
+
32
+ # Or build just this package (after the @ic402/client workspace dep is built):
33
+ pnpm --filter @ic402/mcp build
34
+ ```
35
+
36
+ Run it directly over stdio:
37
+
38
+ ```bash
39
+ node integrations/mcp/dist/index.js
40
+ # or, after build, the bin:
41
+ ic402-mcp
42
+ ```
43
+
44
+ It speaks MCP JSON-RPC on **stdout** and prints diagnostics (including the security banner) to
45
+ **stderr**, so stdout is never polluted.
46
+
47
+ ### Register with an MCP client
48
+
49
+ Point your MCP client at the built entrypoint, passing operator config via env vars and/or a
50
+ config file (see below). For example:
51
+
52
+ ```json
53
+ {
54
+ "mcpServers": {
55
+ "ic402": {
56
+ "command": "node",
57
+ "args": ["/abs/path/to/ic402/integrations/mcp/dist/index.js"],
58
+ "env": {
59
+ "IC402_MCP_PER_CALL_MAX_ATOMIC": "1000000",
60
+ "IC402_MCP_SESSION_MAX_ATOMIC": "5000000"
61
+ }
62
+ }
63
+ }
64
+ }
65
+ ```
66
+
67
+ You can also pass `--config <path>` (or `IC402_MCP_CONFIG=<path>`) to load a JSON config file.
68
+
69
+ ## Configuration
70
+
71
+ ### The `configure` tool
72
+
73
+ `configure` must be called before any other tool — every other tool throws
74
+ `Not configured. Call the "configure" tool first.` until an agent and client exist. It connects
75
+ to a canister and (optionally) loads a signing identity:
76
+
77
+ | Param | Type | Default | Purpose |
78
+ |---|---|---|---|
79
+ | `canisterId` | string | — | Principal of the canister to interact with. |
80
+ | `host` | string | `http://localhost:4944` | ICP replica URL. Root key is fetched automatically when the host contains `localhost`. |
81
+ | `network` | string | `icp:1` | CAIP-2 network identifier. |
82
+ | `identityPem` | string? | — | Path to a secp256k1 PEM (SEC1 `EC PRIVATE KEY` or PKCS#8 `PRIVATE KEY`). Falls back to the `ICP_IDENTITY_PEM` env var; anonymous if neither is set. |
83
+ | `ledger` | string? | — | ICRC-2 ledger canister ID for auto-payment (e.g. ckUSDC). |
84
+ | `autoPayment` | boolean | `false` | Allow paid endpoints to auto-approve/pay (opt-in). |
85
+ | `localDev` | boolean | `false` | Allow `http://localhost` / private / loopback fetch targets (local development only). |
86
+ | `perCallMaxAtomic` | string? | — | Per-call max spend in atomic token units. |
87
+ | `sessionMaxAtomic` | string? | — | Cumulative session max spend in atomic token units. |
88
+
89
+ The identity loader surfaces failures loudly: if `identityPem` is supplied but cannot be parsed
90
+ (unsupported PEM, missing secp256k1 OID, wrong key length), `configure` throws rather than
91
+ silently falling back to anonymous.
92
+
93
+ **The four security knobs (`autoPayment`, `localDev`, `perCallMaxAtomic`, `sessionMaxAtomic`)
94
+ are only honored when the operator enabled `allowSecurityChanges` at startup.** Otherwise they
95
+ are *ignored* — the operator/default config stands — and the `configure` result reports which
96
+ fields were dropped, e.g.:
97
+
98
+ ```
99
+ IGNORED security params ["autoPayment","perCallMaxAtomic"] — operator did not enable
100
+ LLM security changes (set IC402_MCP_ALLOW_SECURITY_CHANGES=1 to allow).
101
+ ```
102
+
103
+ This stops a prompt-injected model from raising its own caps or enabling `localDev` / auto-pay
104
+ through the `configure` tool (`resolveSecurityConfig` in `guards.ts`).
105
+
106
+ ### Operator config (out-of-band) — startup only
107
+
108
+ The real security boundary is resolved once, in `main()`, from an **optional JSON config file +
109
+ environment variables — both inputs the LLM cannot influence**. Precedence is:
110
+
111
+ > built-in defaults < config file < env vars (env wins)
112
+
113
+ A config file is loaded from `--config <path>` or `IC402_MCP_CONFIG`. Unparseable amount values
114
+ fall back to the lower-precedence source rather than crashing the server (`resolveOperatorConfig`
115
+ in `guards.ts`).
116
+
117
+ #### Environment flags
118
+
119
+ | Env var | Config file key | Gates |
120
+ |---|---|---|
121
+ | `IC402_MCP_CONFIG` | — | Path to the JSON config file (alt: `--config <path>`). |
122
+ | `IC402_MCP_PER_CALL_MAX_ATOMIC` | `perCallMaxAtomic` | Per-call spend cap in atomic units. Default `1_000_000` (1.00 USDC at 6 decimals). |
123
+ | `IC402_MCP_SESSION_MAX_ATOMIC` | `sessionMaxAtomic` | Cumulative session spend cap. Default `5_000_000` (5.00 USDC). |
124
+ | `IC402_MCP_LOCAL_DEV` | `localDev` | Allow `http://localhost` / private / loopback fetch targets. Default `false`. |
125
+ | `IC402_MCP_AUTO_PAYMENT` | `autoPayment` | Allow paid endpoints to auto-approve/pay. Default `false`. |
126
+ | `IC402_MCP_ALLOW_SECURITY_CHANGES` | `allowSecurityChanges` | If set, the LLM may retune caps / `localDev` / `autoPayment` via the `configure` tool. Default `false`. |
127
+ | `IC402_MCP_ALLOW_DANGEROUS_TOOLS` | `allowDangerousTools` | If set, enables the default-denied dangerous tools `sign_typed_data` and `delete_content`. Default `false`. |
128
+ | `IC402_MCP_ALLOW_ADMIN_TOOLS` | `allowAdminTools` | If set, enables the default-denied state-changing admin tools (`register_service`, `enable_service`, `claim_job`, `submit_job_result`, `upload_content`). Default `false` (SEC-3). |
129
+
130
+ Booleans are true when the value is exactly `"1"` or `"true"`. Amounts are parsed as
131
+ non-negative integer **strings** (a JS number that has already lost precision is rejected — see
132
+ `parseAtomicAmount`).
133
+
134
+ The signing identity PEM path can also come from the `ICP_IDENTITY_PEM` env var (consumed by the
135
+ `configure` tool).
136
+
137
+ ## Tool inventory
138
+
139
+ All tools are registered in `src/index.ts` via `server.tool(...)` (and `adminTool(...)` for the
140
+ controller-gated set).
141
+
142
+ ### Read-only / query tools
143
+
144
+ - **`call`** — a generic read-only escape hatch. It may only invoke **allowlisted** query/read
145
+ methods on the canister. Anything state-changing, signing, payment, or admin is blocked here;
146
+ use the dedicated tool instead. The allowlist and blocklist are detailed under
147
+ [The `call` allowlist & blocklist](#the-call-allowlist--blocklist).
148
+ - **`list_services`** — list available paid services from the canister.
149
+ - **`request_session`** — request a session intent (pricing: `suggestedDeposit`, `costPerCall`)
150
+ *without* opening a session.
151
+ - **`get_session`** — state of an active session (consumed, remaining, voucher count).
152
+ - **`list_sessions`** — list all active sessions managed by this server.
153
+ - **`get_job_result`** — poll for a job result until it completes or times out.
154
+ - **`session_query`** — send a query through an open session (auto-signs a voucher). Each call
155
+ consumes `costPerCall` from the already-escrowed deposit, so it is not separately confirm-gated.
156
+ - **`search`** — call the canister's `search` endpoint (x402 charge flow); returns results or a
157
+ payment requirement.
158
+ - **`dispute_job`** — dispute a job result (for `BuyerConfirm` verification services).
159
+
160
+ ### Payment / session tools (capped + confirmation-gated)
161
+
162
+ These move or settle value, so they enforce the spend caps and require an explicit `confirm:true`
163
+ (when `confirm` is `false` they return a structured `confirmation_required` proposal describing
164
+ amount / recipient / asset / chain):
165
+
166
+ - **`fetch_x402`** — full x402 flow: probe a gated URL → canister signs the payment → retry with
167
+ the payment header. Probe-only when `confirm:false`; signing the proposed payment requires
168
+ `confirm:true`. SSRF-guarded on the URL and every redirect hop.
169
+ - **`fetch_content`** — fetch content from a `ContentDelivery` response. Supports `inline`,
170
+ `httpUrl` (SSRF-guarded `safeFetch`), `assetCanister` (validated canister id + URL-API origin),
171
+ and `canisterQuery` (restricted to the `getChunk` / `getContent` query methods).
172
+ - **`open_session`** — open a streaming micropayment session. ICP uses ICRC-2 escrow; EVM takes
173
+ an `evmTxHash` (and EIP-3009 `authorization`) proving the USDC deposit. The escrow deposit is
174
+ capped + confirmed, and the deposited amount is counted against the cumulative cap.
175
+ - **`close_session`** — settle the consumed amount on-chain and refund the remainder; returns a
176
+ payment receipt. Requires `confirm:true` (the consumed amount was already counted at deposit
177
+ time, so it is not re-charged against the cap).
178
+ - **`submit_request`** — submit a paid service request. If the service demands payment, the
179
+ operator must have opted into `autoPayment` *and* the caller must pass `confirm:true` after
180
+ reviewing the amount. The price is discovered via a **read-only** `quoteServiceRequest` query
181
+ (no nonce minted), and the cap is checked against the **total** (price + ledger fee), not the
182
+ bare price. Free / session-billed services (total `0`) submit directly.
183
+ - **`register_agent`** — register the canister as an ERC-8004 agent on Base: get nonce + gas →
184
+ canister signs → broadcast → poll receipt. Broadcasting spends native gas, so it is
185
+ confirm-gated (no USDC amount, hence no spend cap); a custom `rpcUrl` is SSRF-validated.
186
+
187
+ ### State-changing / signing tools (the 7 operator tools)
188
+
189
+ Registered via `adminTool(...)`, each binds a **fixed** controller-gated canister method (the LLM
190
+ cannot choose the method) and requires `confirm:true`:
191
+
192
+ | Tool | Canister method | Notes |
193
+ |---|---|---|
194
+ | `upload_content` | `uploadContent` | Upload + encrypt content into the ContentStore (controller-gated). |
195
+ | `delete_content` | `deleteContent` | Destructive delete. **DEFAULT-DENIED** — see below. |
196
+ | `register_service` | `registerService` | Register a paid marketplace service (controller-gated). |
197
+ | `enable_service` | `enableService` | Enable a registered service so it can accept paid requests. |
198
+ | `claim_job` | `claimJob` | Operator claims a pending marketplace job. |
199
+ | `submit_job_result` | `submitJobResult` | Operator submits a job result (+ optional proof) for verification + settlement. |
200
+ | `sign_typed_data` | `signTypedData` | Sign arbitrary EIP-712 typed data with the canister tECDSA key. **DEFAULT-DENIED** — generic signing primitive. |
201
+
202
+ **Default-denied tools (operator opt-in required).** Two categories throw unless the operator
203
+ enables them at startup — a prompt-injected LLM must not reach them via an in-band `confirm` alone:
204
+
205
+ - **Dangerous primitives** — `sign_typed_data` (a raw EIP-712 signing oracle that can authorize an
206
+ arbitrary-value transfer and *bypasses the spend caps*) and `delete_content` (irreversible delete).
207
+ Enable with `IC402_MCP_ALLOW_DANGEROUS_TOOLS=1`.
208
+ - **State-changing admin tools (SEC-3)** — `register_service`, `enable_service`, `claim_job`,
209
+ `submit_job_result`, `upload_content` mutate the canister and drive the value-moving job lifecycle.
210
+ Enable with `IC402_MCP_ALLOW_ADMIN_TOOLS=1`.
211
+
212
+ (`isToolAllowed` / `DANGEROUS_TOOLS` / `ADMIN_TOOLS` in `guards.ts`.)
213
+
214
+ ## Security model
215
+
216
+ The server assumes the LLM driving it may be steered by untrusted web content while it holds a
217
+ controller identity that can sign value transfers from the canister's own EVM address. Every
218
+ money-moving or signing path is therefore capped, SSRF-guarded, and gated behind explicit
219
+ confirmation.
220
+
221
+ ### Spend caps — per-call and cumulative
222
+
223
+ Two caps, in atomic token units (USDC has 6 decimals, so `1_000_000` atomic = 1.00 USDC):
224
+
225
+ - **`perCallMaxAtomic`** — caps a single signed transfer (default `1_000_000` = 1.00 USDC).
226
+ - **`sessionMaxAtomic`** — caps the cumulative total spent/approved across this server session
227
+ (default `5_000_000` = 5.00 USDC).
228
+
229
+ `checkSpend` (in `guards.ts`) rejects negative amounts, any single amount over the per-call cap,
230
+ and any amount that would push the running session total over the cumulative cap. The running
231
+ total (`sessionSpentAtomic`) is only incremented by `commitSpend` **after** a spend actually
232
+ succeeds. The conservative defaults force the LLM to stay within a trivial budget unless a human
233
+ raises the caps out-of-band (and only when `allowSecurityChanges` is set).
234
+
235
+ ### Explicit confirmation gating
236
+
237
+ Money-moving / signing tools route through `requireConfirmation`, which **enforces the caps
238
+ first** (so an over-cap action is refused even before any confirmation) and then, when
239
+ `confirm:false`, returns a structured `confirmation_required` result echoing the proposed
240
+ `amount` / `recipient` / `asset` / `chain` plus the current caps and instructs the caller to
241
+ re-invoke with `confirm:true`. The instruction explicitly says *"Do NOT confirm automatically."*
242
+
243
+ ### Operator-only escalation
244
+
245
+ The security boundary is resolved at startup from a config file + env vars the LLM cannot touch
246
+ (see [Operator config](#operator-config-out-of-band--startup-only)). The `configure` tool can
247
+ loosen caps / `localDev` / `autoPayment` **only** when `allowSecurityChanges` is enabled;
248
+ otherwise those request fields are ignored and reported. The two dangerous tools stay off unless
249
+ `allowDangerousTools` is enabled.
250
+
251
+ ### SSRF guard (and a known limitation)
252
+
253
+ Every outbound fetch path (`fetch_x402`, the `httpUrl` / `assetCanister` deliveries in
254
+ `fetch_content`, and any custom `rpcUrl`) is validated by `validateFetchUrl` and uses the
255
+ redirect-safe `safeFetch` (both in `security.ts`):
256
+
257
+ - Requires `https:` (or `http://localhost` / `127.0.0.1` / `::1` only when `localDev` is set).
258
+ - Rejects literal private / loopback / link-local / CGNAT IPv4 ranges (10/8, 172.16/12,
259
+ 192.168/16, 127/8, 169.254/16, 100.64/10, 0/8), the `169.254.169.254` cloud-metadata endpoint,
260
+ private/loopback/link-local IPv6 (including IPv4-mapped `::ffff:` forms in both dotted and
261
+ hex-compressed notation), and `.local` / `.internal` TLDs.
262
+ - `safeFetch` follows redirects **manually** (default max 5) and **re-validates every hop's
263
+ `Location` before following it**, closing the TOCTOU redirect-bypass where an allowlisted
264
+ origin 30x's to an internal/metadata target. The x402 probe uses the same per-hop
265
+ `validateRedirect` check.
266
+
267
+ > **Known limitation — DNS rebinding.** `validateFetchUrl` validates the *literal* host only. A
268
+ > public hostname that resolves to a private IP is **not** covered. Closing it requires resolving
269
+ > the host and re-checking the resolved addresses before connecting (a resolver seam, noted in
270
+ > `security.ts`).
271
+
272
+ ### The `call` allowlist & blocklist
273
+
274
+ The generic `call` tool is restricted to read-only methods by `isCallMethodAllowed` in
275
+ `index.ts`:
276
+
277
+ 1. **Curated read-only allowlist (authoritative):** `listContent`, `getChunk`, `getAgentCard`,
278
+ `getAgentId`, `verifyGrant`, `listServices`, `getJobStatus`, `getJob`, `getJobResult`,
279
+ `keccak256`, `getPolicyConfig`. These are accepted even if their name contains a blocked
280
+ substring (`getPolicyConfig` contains `policy`).
281
+ 2. **Hard blocklist of substrings** that denote signing/admin/value-moving methods (each has a
282
+ dedicated tool): `sign`, `set`, `submit`, `open`, `close`, `transfer`, `register`, `pay`,
283
+ `approve`, `policy`, `upload`, `delete`, `claim`, `confirm`, `dispute`, `enable`, `disable`,
284
+ `end`. Any method whose name contains one of these is rejected.
285
+ 3. **Forward-compat fallback:** after the blocklist clears a name, read-only getter prefixes
286
+ (`get…`, `list…`, `fetch…`, `is…`) are allowed. Anything else is rejected with a message
287
+ pointing the LLM at the dedicated, capped, confirmation-gated tool.
288
+
289
+ ### Startup security banner
290
+
291
+ On startup the server prints the **effective** security posture to **stderr** (stdout is the MCP
292
+ JSON-RPC channel) so the operator can verify what's active before the agent connects. The banner
293
+ shows the config source (config file path or `env vars / defaults`), `perCallMaxAtomic`,
294
+ `sessionMaxAtomic`, `localDev`, `autoPayment`, `allowSecurityChanges`, and `allowDangerousTools`,
295
+ and warns when any loosened knob is enabled:
296
+
297
+ ```
298
+ ───────────────────────────────────────────────────────────────
299
+ ic402 MCP — effective security config (operator-set, out-of-band)
300
+ source : env vars / defaults
301
+ perCallMaxAtomic : 1000000
302
+ sessionMaxAtomic : 5000000
303
+ localDev : false
304
+ autoPayment : false
305
+ allowSecurityChanges : false (LLM may retune caps via "configure")
306
+ allowDangerousTools : false (sign_typed_data / delete_content)
307
+ ───────────────────────────────────────────────────────────────
308
+ ```
309
+
310
+ When `allowSecurityChanges` or `allowDangerousTools` is enabled, the banner adds:
311
+
312
+ ```
313
+ ⚠ a loosened knob is enabled — only do this in a TRUSTED context (no prompt-injection risk).
314
+ ```
315
+
316
+ ## Source layout
317
+
318
+ - `src/index.ts` — `McpServer` registration, all tool definitions, the `configure` flow, the
319
+ confirmation gate, the `call` allowlist/blocklist, the `adminTool` helper, and `main()`
320
+ (operator-config resolution + security banner + stdio transport).
321
+ - `src/guards.ts` — pure, unit-tested guards: `parseAtomicAmount`, `checkSpend`,
322
+ `resolveSecurityConfig`, `isToolAllowed` / `DANGEROUS_TOOLS`, `resolveOperatorConfig`.
323
+ - `src/security.ts` — pure SSRF validation (`validateFetchUrl` and the IPv4/IPv6 helpers) and the
324
+ redirect-safe `safeFetch`.
325
+ - `package.json` — `@ic402/mcp` v2.1.1; `build` runs `tsc`, `start` runs `node dist/index.js`,
326
+ and the `ic402-mcp` bin maps to `dist/index.js`.
@@ -0,0 +1,79 @@
1
+ /**
2
+ * Spend caps, config-escalation guard, dangerous-tool gate, and safe amount parsing
3
+ * for the ic402 MCP server.
4
+ *
5
+ * Extracted from index.ts so the security-critical decisions are pure and unit-testable
6
+ * (see test/mcp-guards.test.ts). The MCP is driven by a possibly prompt-injected LLM
7
+ * while holding a controller identity, so these guards are the last line between the
8
+ * model and value movement.
9
+ */
10
+ export interface SpendCaps {
11
+ perCallMaxAtomic: bigint;
12
+ sessionMaxAtomic: bigint;
13
+ }
14
+ export interface SecurityConfig extends SpendCaps {
15
+ localDev: boolean;
16
+ autoPayment: boolean;
17
+ }
18
+ export interface ConfigureRequest {
19
+ localDev?: boolean;
20
+ autoPayment?: boolean;
21
+ perCallMaxAtomic?: string;
22
+ sessionMaxAtomic?: string;
23
+ }
24
+ /**
25
+ * Parse a token amount (atomic units) from untrusted input into a bigint, REJECTING
26
+ * floats and JS numbers that have already lost precision. C5: `value: z.number()`
27
+ * silently truncated uint256 amounts above 2^53; callers must pass amounts as decimal
28
+ * strings, and an unsafe number is rejected (not silently used).
29
+ */
30
+ export declare function parseAtomicAmount(input: unknown, field?: string): bigint;
31
+ /**
32
+ * Enforce the per-call and cumulative session spend caps. Throws on violation; never
33
+ * mutates state. (S1/S9 — the cap logic that every value-moving tool must pass.)
34
+ */
35
+ export declare function checkSpend(amountAtomic: bigint, caps: SpendCaps, sessionSpentAtomic: bigint): void;
36
+ /**
37
+ * S8: Resolve the effective security config from an LLM-supplied `configure` request.
38
+ * The caps / localDev / autoPayment knobs loosen the server's security posture, so the
39
+ * LLM may only change them when the OPERATOR opted in at startup (allowSecurityChanges).
40
+ * Otherwise the request's security fields are IGNORED and the operator/default config
41
+ * stands — a prompt-injected model cannot raise its own caps or enable localDev.
42
+ * Returns the resolved config and the list of fields that were ignored (for reporting).
43
+ */
44
+ export declare function resolveSecurityConfig(base: SecurityConfig, req: ConfigureRequest, allowSecurityChanges: boolean): {
45
+ config: SecurityConfig;
46
+ ignored: string[];
47
+ };
48
+ export declare function isToolAllowed(toolName: string, allowDangerousTools: boolean, allowAdminTools: boolean): boolean;
49
+ export interface OperatorConfig {
50
+ security: SecurityConfig;
51
+ allowSecurityChanges: boolean;
52
+ allowDangerousTools: boolean;
53
+ allowAdminTools: boolean;
54
+ }
55
+ /**
56
+ * Resolve the operator's startup security config from an optional JSON config file and the
57
+ * environment. BOTH are out-of-band inputs the LLM cannot influence, so the security boundary
58
+ * stays with the operator (audit S8). Precedence: built-in defaults < config file < env vars
59
+ * (env wins, so an operator can override a file value at launch). Unparseable values fall back
60
+ * to the lower-precedence source rather than throwing, so a typo can't crash the server.
61
+ */
62
+ export declare function resolveOperatorConfig(file: Record<string, unknown> | null, env: Record<string, string | undefined>, defaults: {
63
+ perCallMaxAtomic: bigint;
64
+ sessionMaxAtomic: bigint;
65
+ }): OperatorConfig;
66
+ /** Curated read-only allowlist — every entry is a `query` (or otherwise non-state-changing read)
67
+ * in the IDL. Authoritative: wins over the substring blocklist, so a genuine read-only getter
68
+ * whose name contains a blocked substring (e.g. getPolicyConfig contains 'policy') is still
69
+ * admitted. ONLY add verified non-state-changing query methods. */
70
+ export declare const READONLY_CALL_ALLOWLIST: Set<string>;
71
+ /** Substrings that must NEVER be reachable through the generic `call` path — signing/admin/
72
+ * value-moving method names that have dedicated tools. */
73
+ export declare const CALL_BLOCK_SUBSTRINGS: string[];
74
+ /** Decide whether a method name may be called through the generic (uncapped, unconfirmed) `call`
75
+ * tool. Order: allowlist (authoritative) → substring blocklist → read-only name-prefix fallback. */
76
+ export declare function isCallMethodAllowed(method: string): {
77
+ ok: boolean;
78
+ reason?: string;
79
+ };