Skip to content

Supercharge your AI agents with data from the web and beyond. Building the library for superintelligence. 🔥

feat(mcp): accept error_message and target in the MCP action-log ingest (#5502)

claude[bot]5246476e883f
diff --git a/apps/api/src/services/mcp/action-logs.test.ts b/apps/api/src/services/mcp/action-logs.test.ts
index e5fbda8aa..be30bd861 100644
--- a/apps/api/src/services/mcp/action-logs.test.ts
+++ b/apps/api/src/services/mcp/action-logs.test.ts
@@ -205,6 +205,108 @@ describe("MCP action log contract", () => {
);
});
+ it("stores an optional error message and target, redacted and capped", () => {
+ expect(input()).toMatchObject({ error_message: null, target: null });
+ expect(
+ input({
+ status: "error",
+ error_message: "Request failed with status code 408: Scrape timed out",
+ target: "https://example.com/pricing",
+ }),
+ ).toMatchObject({
+ error_message: "Request failed with status code 408: Scrape timed out",
+ target: "https://example.com/pricing",
+ });
+ expect(input({ error_message: "", target: " " })).toMatchObject({
+ error_message: null,
+ target: null,
+ });
+ expect(
+ input({ error_message: "Unauthorized:\nBearer abc.def for fc-123abc" })
+ .error_message,
+ ).toBe("Unauthorized: [redacted] for [redacted]");
+ expect(
+ input({ target: "https://example.com/?key=fc-123abc&page=2" }).target,
+ ).toBe("https://example.com/?key=[redacted]&page=[redacted]");
+ expect(
+ input({
+ target:
+ "https://user:[email protected]/oauth/callback?code=c0de&state=s7ate&flag#access_token=t",
+ }).target,
+ ).toBe(
+ "https://[redacted]@example.com/oauth/callback?code=[redacted]&state=[redacted]&[redacted]#[redacted]",
+ );
+ expect(
+ input({
+ target:
+ "https://files.example.com/scl/fi/k8sj2h4l9q0w1e3r5t7y/report.pdf?dl=0",
+ }).target,
+ ).toBe(
+ "https://files.example.com/scl/fi/[redacted]/report.pdf?dl=[redacted]",
+ );
+ expect(
+ input({ target: "https://example.com/reset-password/abc" }).target,
+ ).toBe("https://example.com/reset-password/[redacted]");
+ expect(
+ input({ target: "https://example.com/blog/how-to-build-an-api-in-2024" })
+ .target,
+ ).toBe("https://example.com/blog/how-to-build-an-api-in-2024");
+ expect(
+ input({
+ error_message:
+ "Authorization: Basic dXNlcjpwYXNz; otp=123456 resetToken: r3s3t verification_code=99 bad ?code=abc&state=def",
+ }).error_message,
+ ).toBe(
+ "Authorization: [redacted]; otp=[redacted] resetToken: [redacted] verification_code=[redacted] bad ?code=[redacted]&state=[redacted]",
+ );
+ expect(input({ error_message: "Status code: 500" }).error_message).toBe(
+ "Status code: 500",
+ );
+ expect(
+ input({
+ error_message:
+ "{'password': 'hunter2', 'api_key': 'abc123secret'} {'Authorization': 'Basic dXNlcjpwYXNz'} {\"token\"=>\"t0k\"}",
+ }).error_message,
+ ).toBe(
+ "{'password': [redacted], 'api_key': [redacted]} {'Authorization': [redacted]} {\"token\"=>[redacted]}",
+ );
+ expect(
+ input({ error_message: "401 for Basic dXNlcjpwYXNz on Basic research" })
+ .error_message,
+ ).toBe("401 for [redacted] on Basic research");
+ expect(
+ input({
+ error_message: "password=>Basic dXNlcjpwYXNz cred: Basic dXNlcjpwYXNz",
+ }).error_message,
+ ).toBe("password=>[redacted] cred: [redacted]");
+ expect(input({ error_message: "401 for Basic YTo=" }).error_message).toBe(
+ "401 for [redacted]",
+ );
+ for (const [message, expected] of [
+ ["password=Basic dXNlcjpwYXNz", "password=[redacted]"],
+ ["x-auth: Basic dXNlcjpwYXNz", "x-auth: [redacted]"],
+ ["password=Basic c2VjcmV0dG9rZW4", "password=[redacted]"],
+ ["auth_token: Token abc123tok", "auth_token: [redacted]"],
+ ["secret=Bearer abc123tok", "secret=[redacted]"],
+ ["password=hunter2 Basic research", "password=[redacted] Basic research"],
+ ]) {
+ expect(input({ error_message: message }).error_message).toBe(expected);
+ }
+ expect(input({ error_message: "x".repeat(400) }).error_message).toBe(
+ "x".repeat(300),
+ );
+ expect(input({ target: "q".repeat(600) }).target).toBe("q".repeat(512));
+ expect(input({ target: `${"q".repeat(511)}\u{1F600}` }).target).toBe(
+ "q".repeat(511),
+ );
+ expect(() => input({ error_message: 500 })).toThrow(
+ "error_message must be a string",
+ );
+ expect(() => input({ target: ["https://example.com"] })).toThrow(
+ "target must be a string",
+ );
+ });
+
it("stores an optional coarse error code and rejects anything but snake_case", () => {
expect(input().error_code).toBeNull();
expect(
@@ -517,6 +619,8 @@ describe("MCP action log contract", () => {
objective: "Find the refund policy",
api_job_id: REQUEST_ID,
error_code: "rate_limited",
+ error_message: "Rate limit exceeded",
+ target: "https://example.com/pricing",
user_agent: "Claude/1.0",
mcp_origin: "mcp-ua-codex-cli-rs",
mcp_client: "codex-cli-rs",
@@ -541,6 +645,8 @@ describe("MCP action log contract", () => {
objective: "Find the refund policy",
api_job_id: REQUEST_ID,
error_code: "rate_limited",
+ error_message: "Rate limit exceeded",
+ target: "https://example.com/pricing",
mcp_origin: "mcp-ua-codex-cli-rs",
mcp_client: "codex-cli-rs",
mcp_client_source: "user_agent",
@@ -651,7 +757,7 @@ describe("MCP action log contract", () => {
expect(complete.nextCursor).toBeNull();
});
- it("never selects the objective or API job ID for the team-readable list", async () => {
+ it("never selects call content or the API job ID for the team-readable list", async () => {
ch.client = { query: vi.fn().mockResolvedValue({ json: async () => [] }) };
await listMcpActionLogs(TEAM_ID);
const sql = ch.client.query.mock.calls[0][0].query as string;
@@ -659,6 +765,8 @@ describe("MCP action log contract", () => {
expect(columns).toContain("tool_name");
expect(columns).not.toContain("objective");
expect(columns).not.toContain("api_job_id");
+ expect(columns).not.toContain("error_message");
+ expect(columns).not.toMatch(/\btarget\b/);
});
it("reports storage as unavailable when the read fails", async () => {
diff --git a/apps/api/src/services/mcp/action-logs.ts b/apps/api/src/services/mcp/action-logs.ts
index 24cafd89c..fa80abf80 100644
--- a/apps/api/src/services/mcp/action-logs.ts
+++ b/apps/api/src/services/mcp/action-logs.ts
@@ -18,7 +18,23 @@ const POSTGRES_BIGINT_MAX = 9_223_372_036_854_775_807n;
const SECRET_PATTERN =
/(?:\bBearer\s+\S+|\beyJ[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\b|\b(?:sk-|fc-|fco_|fcr_|fcmcp_)[A-Za-z0-9_-]+)/i;
const SECRET_PATTERN_ALL = new RegExp(SECRET_PATTERN.source, "gi");
+// After any secret key, a leading auth scheme (Basic, Bearer, Token, ...) is
+// consumed with its credential so no part of "password=>Basic <b64>" survives.
+const DETAIL_SECRET_PAIR = new RegExp(
+ "(\\b(?:[A-Za-z0-9]*[_.-]?)(?:api[_-]?key|apikey|key|access[_-]?token|accesstoken|auth[_-]?token|token|client[_-]?secret|secret|password|passwd|passcode|pwd|auth|authorization|signature|sig|credentials?|session[_-]?id|sessionid|otp|nonce|verifier|cookie|jwt|(?:auth|oauth|authorization|verification|confirmation|reset|access|one[_-]?time)[_-]?code)[\"']?)(\\s*(?:=>|[=:])\\s*)(?:\"[^\"]*\"|'[^']*'|(?:(?:(?<=authorization[\"']?\\s*(?:=>|[=:])\\s*)[A-Za-z]+|Basic|Bearer|Token|Digest|Bot|ApiKey|Negotiate|NTLM)\\s+)?[^\\s&#\"',;]+)",
+ "gi",
+);
+// A Basic credential is base64 of "user:password"; only one that decodes
+// to a colon pair is redacted, so prose such as "Basic research" is kept.
+const DETAIL_BASIC_AUTH = /\bBasic\s+([A-Za-z0-9+/]{2,}={0,2})/gi;
+const DETAIL_QUERY_PAIR = /([?&][^\s=&#?]+=)[^\s&#]+/g;
+const DETAIL_URL = /\b[a-z][a-z0-9+.-]*:\/\/[^\s"'<>`]+/gi;
+const DETAIL_TOKEN_RUN = /[A-Za-z0-9]{16,}/g;
+const DETAIL_TOKEN_PATH =
+ /^(?:s|share|shared|invite|invitation|reset|reset[_-]?password|password[_-]?reset|verify|verification|confirm|confirmation|activate|magic|magic[_-]?link|token|auth|login|unsubscribe|callback|otp)$/i;
const OBJECTIVE_MAX_LENGTH = 1000;
+const ERROR_MESSAGE_MAX_LENGTH = 300;
+const TARGET_MAX_LENGTH = 512;
const ERROR_CODE_PATTERN = /^(?=.{2,64}$)[a-z][a-z0-9]*(?:_[a-z0-9]+)*$/;
const MCP_CLIENT_SOURCES = ["client_info", "user_agent", "oauth"] as const;
const RATE_LIMIT_RETRY_OUTCOMES = [
@@ -51,6 +67,8 @@ const ALLOWED_FIELDS = new Set([
"client_version",
"error_class",
"error_code",
+ "error_message",
+ "target",
"resource",
"objective",
"api_job_id",
@@ -125,16 +143,95 @@ function errorCode(value: unknown) {
return value;
}
-function objective(value: unknown) {
+function tokenLikeSegment(segment: string) {
+ return (segment.match(DETAIL_TOKEN_RUN) ?? []).some(
+ run => /\d/.test(run) && /[A-Za-z]/.test(run),
+ );
+}
+
+/**
+ * Keeps a URL's scheme, host and plain path: userinfo, every query value, the
+ * fragment and token-like path segments become [redacted].
+ */
+function scrubbedUrl(url: string) {
+ const hashAt = url.indexOf("#");
+ const fragment =
+ hashAt === -1 || hashAt === url.length - 1 ? "" : "#[redacted]";
+ const beforeHash = hashAt === -1 ? url : url.slice(0, hashAt);
+ const queryAt = beforeHash.indexOf("?");
+ const base = queryAt === -1 ? beforeHash : beforeHash.slice(0, queryAt);
+ const query =
+ queryAt === -1
+ ? ""
+ : "?" +
+ beforeHash
+ .slice(queryAt + 1)
+ .split("&")
+ .map(pair => {
+ const eq = pair.indexOf("=");
+ if (eq === -1) return pair && "[redacted]";
+ return eq === pair.length - 1
+ ? pair
+ : pair.slice(0, eq + 1) + "[redacted]";
+ })
+ .join("&");
+ const match = /^([^:]+:\/\/)([^/]*)(.*)$/.exec(base);
+ if (!match) return base + query + fragment;
+ const [, scheme, authority, path] = match;
+ const host = authority.includes("@")
+ ? "[redacted]" + authority.slice(authority.lastIndexOf("@"))
+ : authority;
+ const segments = path.split("/");
+ const scrubbedPath = segments
+ .map((segment, i) =>
+ tokenLikeSegment(segment) ||
+ (i > 0 && DETAIL_TOKEN_PATH.test(segments[i - 1]))
+ ? "[redacted]"
+ : segment,
+ )
+ .join("/");
+ return scheme + host + scrubbedPath + query + fragment;
+}
+
+/**
+ * Error messages and targets also lose URL userinfo, query values, fragments,
+ * token-like path segments, bare query-string values, secret=value pairs
+ * (keys bare or quoted, as in JSON or Python and Ruby dicts) and Basic
+ * credentials.
+ */
+function scrubbedDetail(text: string) {
+ return text
+ .replace(SECRET_PATTERN_ALL, "[redacted]")
+ .replace(DETAIL_URL, scrubbedUrl)
+ .replace(DETAIL_QUERY_PAIR, "$1[redacted]")
+ .replace(DETAIL_BASIC_AUTH, (match, token: string) =>
+ Buffer.from(token, "base64").toString("latin1").includes(":")
+ ? "[redacted]"
+ : match,
+ )
+ .replace(DETAIL_SECRET_PAIR, "$1$2[redacted]");
+}
+
+/**
+ * Free text from the call (objective, error message, target URL):
+ * control characters become spaces, secret-like values become [redacted],
+ * and the result is cut to `max` characters rather than rejected.
+ */
+function redactedText(
+ value: unknown,
+ field: string,
+ max: number,
+ scrub: (text: string) => string = text => text,
+) {
if (value == null) return null;
- if (typeof value !== "string") invalid("objective must be a string");
- const normalized = value
- .slice(0, OBJECTIVE_MAX_LENGTH * 4)
- .replace(/[\u0000-\u001F\u007F-\u009F]+/g, " ")
+ if (typeof value !== "string") invalid(`${field} must be a string`);
+ const normalized = scrub(
+ value.slice(0, max * 4).replace(/[\u0000-\u001F\u007F-\u009F]+/g, " "),
+ )
.replace(SECRET_PATTERN_ALL, "[redacted]")
.replace(/\beyJ[A-Za-z0-9_.-]*$/, "[redacted]")
.trim()
- .slice(0, OBJECTIVE_MAX_LENGTH)
+ .slice(0, max)
.replace(/[\uD800-\uDBFF]$/, "")
.trim();
return normalized || null;
@@ -211,6 +308,8 @@ type McpActionLogInput = {
client_version: string | null;
error_class: string | null;
error_code: string | null;
+ error_message: string | null;
+ target: string | null;
resource: string;
objective: string | null;
api_job_id: string | null;
@@ -311,8 +410,24 @@ export function normalizeMcpActionLogInput(
client_version: text(payload.client_version, "client_version"),
error_class: text(payload.error_class, "error_class"),
error_code: errorCode(payload.error_code),
+ error_message: redactedText(
+ payload.error_message,
+ "error_message",
+ ERROR_MESSAGE_MAX_LENGTH,
+ scrubbedDetail,
+ ),
+ target: redactedText(
+ payload.target,
+ "target",
+ TARGET_MAX_LENGTH,
+ scrubbedDetail,
+ ),
resource,
- objective: objective(payload.objective),
+ objective: redactedText(
+ payload.objective,
+ "objective",
+ OBJECTIVE_MAX_LENGTH,
+ ),
api_job_id: text(payload.api_job_id, "api_job_id"),
mcp_origin: text(payload.mcp_origin, "mcp_origin"),
mcp_client: text(payload.mcp_client, "mcp_client"),
@@ -481,8 +596,8 @@ export async function listMcpActionLogs(
let rows: McpActionLogRow[];
try {
const result = await clickhouseClient.query({
- // The explicit column list is the team-readable contract: objective and
- // api_job_id stay out of it. Filters name `log.` columns because the
+ // The explicit column list is the team-readable contract: objective,
+ // api_job_id, error_message and target stay out of it. Filters name `log.` columns because the
// formatted created_at alias would otherwise shadow the stored value.
query: `SELECT id, user_id, toString(log.api_key_id) AS api_key_id, oauth_client_id, auth_type, tool_name, status, request_id, client_name, client_version, error_class, resource, formatDateTime(log.created_at, '%Y-%m-%dT%H:%i:%S.%fZ', 'UTC') AS created_at FROM mcp_action_logs AS log FINAL WHERE log.team_id = {teamId: UUID} AND log.created_at > now64(9) - INTERVAL ${RETENTION_DAYS} DAY${cursor ? " AND (log.created_at, log.id) < (parseDateTime64BestEffort({before: String}, 9), {beforeId: UUID})" : ""} ORDER BY log.created_at DESC, log.id DESC LIMIT {limit: UInt32}`,
query_params: {