Skip to content

Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure

TypeScript

fix(mcp-host): keep Host non-chat JSON on the 10mb parser

Alfredo Lopez2e4179042733
diff --git a/docs/architecture/codex-subscription-transport-contract.md b/docs/architecture/codex-subscription-transport-contract.md
index eb74e672b..811f2f205 100644
--- a/docs/architecture/codex-subscription-transport-contract.md
+++ b/docs/architecture/codex-subscription-transport-contract.md
@@ -62,8 +62,9 @@ the shared visual envelope budget refuses to start.
Desktop enforces the 16 MiB individual and 16 MiB combined hard attachment
budgets (usual target remains 5 / 9 MiB).
-RPC and Host permit a 24 MiB JSON body only on their message POST routes; their
-6 MiB budget for other content and routes remains unchanged. The proxy's larger
+RPC and Host permit a 24 MiB JSON body only on their message POST routes; chat
+non-image bytes stay on the 6 MiB share. Non-chat rpc-proxy and Host control
+routes keep the 10 MB ordinary JSON cap. The proxy's larger
parser requires a valid platform identity on the visual completion route.
Admin and unauthenticated requests retain the ordinary configured body limit.
`CODEX_LLM_PROXY_MAX_VISUAL_BODY_BYTES` controls the visual transport ceiling;
diff --git a/docs/testing/codex-subscription-validation.md b/docs/testing/codex-subscription-validation.md
index af321107b..b74549aba 100644
--- a/docs/testing/codex-subscription-validation.md
+++ b/docs/testing/codex-subscription-validation.md
@@ -14,7 +14,7 @@ Capacity checks cover the usual 5 / 9 / 14 MiB target and the hard 16 MiB
per-image / 16 MiB aggregate / 24 MiB envelope ceiling, plus the 2048 px
model bound. Exercise 5 MiB, 12 MiB (exceptional, above 10 MiB), 10+5 MiB
and 3x5 MiB acceptance, plus overflow above 16 MiB. V1 and V2 non-image fields keep a 1 MiB ceiling;
-unrelated RPC/Host routes keep 6 MiB. The larger proxy parser must not admit an
+chat non-image bytes stay at 6 MiB; unrelated rpc-proxy and Host control routes keep the 10 MB ordinary JSON cap. The larger proxy parser must not admit an
anonymous, wrong-scope or admin request, or enlarge V1's effective body limit.
Changing these budgets does not change OAuth, grants, ticket binding, origins,
fallback, or connection selection. Local capacity evidence is separate from
diff --git a/mcp-host/package-lock.json b/mcp-host/package-lock.json
index a366bc26f..cd947ff9e 100644
--- a/mcp-host/package-lock.json
+++ b/mcp-host/package-lock.json
@@ -1,12 +1,12 @@
{
"name": "clerum-mcp-host",
- "version": "0.1.367",
+ "version": "0.1.368",
"lockfileVersion": 3,
"requires": true,
"packages": {
"": {
"name": "clerum-mcp-host",
- "version": "0.1.367",
+ "version": "0.1.368",
"license": "SEE LICENSE IN LICENSE",
"dependencies": {
"@anthropic-ai/sdk": "^0.32.0",
diff --git a/mcp-host/package.json b/mcp-host/package.json
index 957c51bd2..3dc925d7b 100644
--- a/mcp-host/package.json
+++ b/mcp-host/package.json
@@ -1,6 +1,6 @@
{
"name": "clerum-mcp-host",
- "version": "0.1.367",
+ "version": "0.1.368",
"description": "MCP Host - reads Host CRD and provides LLM access via OpenAI/Claude",
"main": "dist/main.js",
"engines": {
diff --git a/mcp-host/src/__tests__/bodylimits.test.ts b/mcp-host/src/__tests__/bodylimits.test.ts
index 51a2ace6b..22471a7dd 100644
--- a/mcp-host/src/__tests__/bodylimits.test.ts
+++ b/mcp-host/src/__tests__/bodylimits.test.ts
@@ -8,7 +8,8 @@
* - The attachments arrive byte-identical: nothing is dropped, truncated or
* re-encoded on the way through the parser.
* - The larger ceiling is NOT a general text allowance: non-image bytes stay
- * capped at 6MiB, and every other route keeps its 6MiB parser.
+ * capped at 6MiB. Every other Host route keeps the 10mb ordinary JSON
+ * parser, matching rpc-proxy `jsonBody`.
*
* The auth boundary is exercised in its documented "auth disabled" test mode
* (`CLERUM_ENABLE_AUTH=false`), matching server.attachments.test.ts; the edge
@@ -329,12 +330,19 @@ describe('mcp-host runtime message body budget', () => {
expect(captured).toHaveLength(0)
})
- it('keeps the 6MiB parser on every other POST route', async () => {
- const response = await fetch(`${baseUrl}/v1/runtime/model`, {
+ it('keeps the 10mb parser on every other POST route', async () => {
+ const over = await fetch(`${baseUrl}/v1/runtime/model`, {
+ method: 'POST',
+ headers: rpcProxyEdgeHeaders(),
+ body: JSON.stringify({ model: 'x'.repeat(11 * MIB) }),
+ })
+ expect(over.status).toBe(413)
+
+ const under = await fetch(`${baseUrl}/v1/runtime/model`, {
method: 'POST',
headers: rpcProxyEdgeHeaders(),
body: JSON.stringify({ model: 'x'.repeat(7 * MIB) }),
})
- expect(response.status).toBe(413)
+ expect(under.status).not.toBe(413)
})
})
diff --git a/mcp-host/src/server.ts b/mcp-host/src/server.ts
index 854de7be3..9d45efb78 100644
--- a/mcp-host/src/server.ts
+++ b/mcp-host/src/server.ts
@@ -308,9 +308,10 @@ export class RPCServer {
})
// Image attachments are sent as base64 in /v1/runtime/messages.
- // That route alone gets the 24 MiB ceiling. Every other route uses this
- // PR's 6mb non-image default (not the historical 10mb parser).
- const jsonParser = express.json({ limit: '6mb' })
+ // That route alone gets the 24 MiB ceiling. Every other Host route uses
+ // the same 10mb ordinary JSON cap as rpc-proxy `jsonBody`, so a future
+ // non-chat control body cannot 413 here and pass the proxy.
+ const jsonParser = express.json({ limit: '10mb' })
const chatJsonParser = express.json({
limit: MAX_CHAT_BODY_BYTES,
verify: (req, _res, buffer) => {