@mikeargento/bitgraph-mcp 0.3.0 → 0.4.0

package/dist/server.js

@@ -2,69 +2,92 @@
 /**
  * @mikeargento/bitgraph-mcp: tool definitions.
  *
- * Three gestures, the same three the website has: make a BitGraph of a file,
- * check whether bytes are on record, fetch a proof. Making a BitGraph builds a
- * fused artifact from the file in memory, on this machine, and commits its
- * digest under a slot allocated for it; only SHA-256 digests and slot records
- * ever leave the machine. File contents are never uploaded.
+ * Three gestures, the same three the website has: make a BitGraph, check
+ * whether bytes are on record, fetch a proof. Making a BitGraph is one
+ * gesture for any number of files: everything in a call becomes a member of
+ * ONE set under ONE slot, the way a drop on the site does. Each file is read
+ * once, on this machine, for its digest and a hasher state; the new fused
+ * bytes are never written and never held, their digest is finished from that
+ * state once the slot exists; and the set's committed artifact is hashed
+ * and committed under the same slot. Only digests, that artifact and slot
+ * records leave the machine. File contents are never uploaded and files are
+ * never modified.
  */
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { z } from "zod";
 import { readFile } from "node:fs/promises";
-import { fuse, builderFor, placementForBytes, fusedNamesFor } from "@mikeargento/bitgraph";
-import { ApiError, batchCheck, configFromEnv, getProofDetail, search, } from "./api.js";
+import { FuseError, MAX_SET_MEMBERS, fuseSet } from "@mikeargento/bitgraph";
+import { ApiError, batchCheck, configFromEnv, getProofDetail, indexSetMembers, search, } from "./api.js";
 import { fromUrlSafeB64, looksLikeDigest, mapConcurrent, sha256FileB64, toUrlSafeB64, } from "./encoding.js";
 import { capJson, positionOf, proofUrl, renderCheckMarkdown, renderProofMarkdown, renderRecordMarkdown, } from "./format.js";
-export const SERVER_VERSION = "0.2.0";
-const HASH_CONCURRENCY = 4;
-const MAX_FILES = 500;
-/** Files above this are refused: the fused artifact is built in memory. */
-const MAX_FUSE_BYTES = 256 * 1024 * 1024;
+import { expandPaths, fusedDigestFor, scanFile } from "./scan.js";
+export const SERVER_VERSION = "0.4.0";
+const SCAN_CONCURRENCY = 4;
+/** Paths per call; a directory counts once and expands to its files. */
+const MAX_PATHS = 2000;
+/** Files one call may BitGraph after directories expand: one set. The site's own ceiling for a set/2. */
+export const MAX_MEMBERS = 100_000;
+/** Files one check may cover after directories expand. */
+const MAX_CHECK_FILES = 10_000;
+/** A file whose length changed while it was read is fused from its bytes instead; above this it is left out rather than held in memory. */
+const MAX_LOADED_BYTES = 256 * 1024 * 1024;
+/** Rows the structured result lists in full; every fused row shares the set's position. */
+export const ROW_CAP = 500;
+/** Members' evidence per set-index request: the site's own chunk. */
+export const SET_INDEX_CHUNK = 2500;
 /**
- * The default pipeline, the same one the site's drop and the bitgraph-fuse
- * command run: choose the placement from the bytes, allocate a slot, derive
- * the commitment, build the fused bytes, hash them, commit under that exact
- * slot, verify the returned proof against the bytes.
+ * The default pipeline, the one the site's drop runs: one slot for the set,
+ * every member a hashed member whose fused digest is finished from the
+ * scan's open hasher with its placement's suffix for that slot, the set's
+ * manifest (or, for a set/2, its root document) committed under the same
+ * slot, and the returned proof verified against the committed artifact with
+ * every member bound to it by digest. A file whose length changed during the
+ * scan is a loaded member: read again when it is its turn, checked against
+ * the scan's digest, fused in memory, hashed and released.
  */
-async function fuseFileDefault(bytes, name, config) {
-    const placement = placementForBytes(bytes);
-    const { fusedName } = fusedNamesFor(name, placement);
-    const r = await fuse(builderFor(placement, bytes), {
-        placement,
-        original: bytes,
-        fusedFile: fusedName,
+async function fuseSetDefault(files, config, opts) {
+    const members = files.map((f) => f.state !== null
+        ? { originDigest: f.originDigest, placement: f.placement, name: f.name, fusedDigest: ({ commitment }) => fusedDigestFor(f, commitment) }
+        : { load: async () => new Uint8Array(await readFile(f.path)), originDigest: f.originDigest, placement: f.placement, name: f.name });
+    const r = await fuseSet(members, {
+        set: opts.set,
         keepFused: false,
+        ...(opts.onProgress !== undefined ? { onProgress: opts.onProgress } : {}),
         transport: { baseUrl: config.baseUrl, ...(config.apiKey ? { apiKey: config.apiKey } : {}) },
     });
     return {
+        set: r.set,
         proof: r.proof,
-        frame: r.frame,
-        placement,
         artifactDigestB64: r.artifactDigestB64,
-        originDigestB64: r.originDigestB64 ?? "",
+        count: r.members.length,
+        manifestEchoed: r.manifestEchoed,
+        recovered: r.recovered,
+        members: r.members.map((m) => ({
+            index: m.index,
+            manifestIndex: m.manifestIndex,
+            placement: m.placement,
+            originDigestB64: m.originDigestB64,
+            artifactDigestB64: m.artifactDigestB64,
+            ...(m.memberProof !== undefined ? { memberProof: m.memberProof } : {}),
+        })),
     };
 }
-/** Read the given paths whole (bounded concurrency). Throws before any network call. */
-async function readPaths(paths) {
+/** Hash the given paths (bounded concurrency). Throws before any network call. */
+async function hashPaths(paths) {
     const failures = [];
-    const out = await mapConcurrent(paths, HASH_CONCURRENCY, async (p) => {
+    const digests = await mapConcurrent(paths, SCAN_CONCURRENCY, async (p) => {
         try {
-            const buf = await readFile(p);
-            if (buf.length > MAX_FUSE_BYTES)
-                throw new Error(`larger than ${MAX_FUSE_BYTES / (1024 * 1024)} MB; the fused artifact is built in memory`);
-            const bytes = new Uint8Array(buf);
-            const digestB64 = await sha256FileB64(p);
-            return { bytes, digestB64 };
+            return await sha256FileB64(p);
         }
         catch (err) {
             failures.push(`${p}: ${err instanceof Error ? err.message : String(err)}`);
-            return { bytes: new Uint8Array(0), digestB64: "" };
+            return "";
         }
     });
     if (failures.length > 0) {
-        throw new Error(`Could not read ${failures.length} file(s); nothing was BitGraphed.\n${failures.join("\n")}\nUse absolute paths to existing regular files under ${MAX_FUSE_BYTES / (1024 * 1024)} MB.`);
+        throw new Error(`Could not read ${failures.length} file(s); nothing was checked.\n${failures.join("\n")}\nUse absolute paths to existing regular files.`);
     }
-    return out;
+    return digests;
 }
 const responseFormatSchema = z
     .enum(["markdown", "json"])
@@ -86,48 +109,106 @@ function errorText(err) {
     }
     return `Error: ${err instanceof Error ? err.message : String(err)}`;
 }
-/** Hash the given paths (bounded concurrency). Throws before any network call. */
-async function hashPaths(paths) {
-    const failures = [];
-    const digests = await mapConcurrent(paths, HASH_CONCURRENCY, async (p) => {
-        try {
-            return await sha256FileB64(p);
+/** Why the set was not made, and what to do next. Never a success-looking line. */
+function setFailureText(err) {
+    if (err instanceof FuseError) {
+        const where = err.member !== null ? ` (member ${err.member})` : "";
+        switch (err.code) {
+            case "tee-restarting":
+                return `Nothing was BitGraphed: ${err.message}. The boundary restarts once a day at 23:59 UTC; run bitgraph_record again with the same paths in a minute.`;
+            case "network":
+            case "transport":
+                return `Nothing is known to be BitGraphed: ${err.message}. Run bitgraph_record again with the same paths: files a set did land come back as on record and are not made again.`;
+            default:
+                return `Nothing was BitGraphed${where}: ${err.message} (${err.code}). Run bitgraph_record again with the same paths.`;
         }
-        catch (err) {
-            failures.push(`${p}: ${err instanceof Error ? err.message : String(err)}`);
-            return "";
+    }
+    return `Nothing was BitGraphed: ${err instanceof Error ? err.message : String(err)}. Run bitgraph_record again with the same paths.`;
+}
+/** Progress notifications, when the client asked for them with a progress token; a no-op otherwise. */
+function progressReporter(extra) {
+    const token = extra._meta?.progressToken;
+    if (token === undefined)
+        return () => { };
+    return (progress, total, message) => {
+        void extra.sendNotification({ method: "notifications/progress", params: { progressToken: token, progress, total, message } }).catch(() => { });
+    };
+}
+const PHASES = {
+    hash: "checking members",
+    fuse: "fusing",
+    tree: "building the tree",
+    commit: "committing",
+    verify: "verifying",
+};
+const pendingIndex = [];
+/** How many members' evidence is waiting to be indexed (tests read it). */
+export function pendingIndexCount() {
+    return pendingIndex.reduce((n, p) => n + p.members.length, 0);
+}
+/** Send pending evidence in chunks; what fails stays pending. */
+async function flushIndex(config, report) {
+    const total = pendingIndexCount();
+    let written = 0;
+    let stopped = false;
+    for (const set of pendingIndex) {
+        while (set.members.length > 0 && !stopped) {
+            const chunk = set.members.slice(0, SET_INDEX_CHUNK);
+            let ok = false;
+            for (let attempt = 0; attempt < 2 && !ok; attempt++) {
+                try {
+                    await indexSetMembers(config, { setDigest: set.setDigest, epoch: set.epoch, counter: set.counter, members: chunk });
+                    ok = true;
+                }
+                catch {
+                    ok = false;
+                }
+            }
+            if (!ok) {
+                stopped = true;
+                break;
+            }
+            set.members.splice(0, chunk.length);
+            written += chunk.length;
+            report(written, total, `indexing ${written} of ${total}`);
         }
-    });
-    if (failures.length > 0) {
-        throw new Error(`Could not read ${failures.length} file(s); nothing was recorded.\n${failures.join("\n")}\nUse absolute paths to existing regular files.`);
+        if (stopped)
+            break;
     }
-    return digests;
+    const left = pendingIndex.filter((s) => s.members.length > 0);
+    pendingIndex.length = 0;
+    pendingIndex.push(...left);
+    return { written, pending: total - written };
 }
 export function buildServer(deps = {}) {
-    const fuseFile = deps.fuseFile ?? fuseFileDefault;
+    const fuseSetPipeline = deps.fuseSet ?? fuseSetDefault;
     const server = new McpServer({
         name: "bitgraph-mcp-server",
         version: SERVER_VERSION,
+    }, {
+        instructions: "BitGraph gives a file's bytes a causal position in a public ledger bracketed by Ethereum anchors. bitgraph_record makes ONE BitGraph of everything in a call, files and folders alike: one set under one slot, one position, every file's new fused bytes listed by digest in the committed artifact. " +
+            "Files are read on this machine and never uploaded or modified; the new bytes are virtual and never written. Recordings are permanent: only make BitGraphs of files the user asked for, and never generate content just to record it. bitgraph_check and bitgraph_get_proof are read-only.",
     });
     server.registerTool("bitgraph_record", {
         title: "Make a BitGraph",
-        description: "Make a BitGraph of one or more files. For each file, on this machine: hash it (the origin), allocate an unused slot in the BitGraph ledger (bitgraph.ing) before any artifact exists, build a new fused artifact from the file in memory with a registered placement (a 48-byte trailer for formats that ignore trailing bytes such as JPEG, PNG, TIFF, WebP; a small tar container otherwise), hash it, and commit that digest under the same slot. " +
-            "The file is never modified and never uploaded; only digests and slot records leave the machine. The fused bytes are not kept: the original plus the proof rebuilds them, and the Frame for each file is returned in the structured result. " +
-            "Files whose bytes are already on record, as a recording or as the origin of a fused artifact, are NOT BitGraphed again by default; they come back as 'on record' with their earliest position. " +
-            "Pass again=true to deliberately make a new fused artifact from a file already on record. " +
-            "BitGraphs are permanent: the ledger has 10-year retention and no deletes, so only BitGraph files the user asked to. " +
-            "Returns one outcome per file: 'fused' (a new fused artifact, with its placement, position and proof page URL), 'on record', or 'not fused' (with the error). " +
-            "Use bitgraph_check instead when the user only wants to know whether a file is on record.",
+        description: "Make a BitGraph of files or folders. Everything in one call becomes ONE BitGraph: a set under a single slot in the BitGraph ledger (bitgraph.ing), one position for all of it, the way a drop on the site works. " +
+            "On this machine each file is read once for its SHA-256 (the origin) and a hasher state; an unused slot is allocated before any new file exists; every file's new fused bytes (the original plus a registered placement carrying the slot's commitment: a 48-byte trailer for JPEG, PNG, GIF, TIFF and TIFF-based raws, BMP, WebP, WAV and AVI, a small tar container with the original first for everything else) are hashed from that state without being written or held; and the canonical list of those digests (above 2,000 files, a Merkle root over it) is committed under the same slot. " +
+            "Files are never modified and never uploaded: only digests, the committed artifact and slot records leave the machine. " +
+            "Give file paths, directory paths, or both (absolute paths preferred): a directory is every regular file under it, recursively, with hidden entries and symbolic links left out. " +
+            "Files already on record (recorded, or the origin of a fused file) are NOT made again by default; they come back as 'on record' with their earliest position. Pass again=true to make a new BitGraph of them deliberately. " +
+            "BitGraphs are permanent: the ledger has 10-year retention and no deletes, so only BitGraph files the user asked to, and never generate content just to record it. " +
+            "Returns the set's position and proof page, and one outcome per file: 'fused' (its row in the set, one of N), 'on record', or 'not fused' (with the reason). A lookup by any file's own digest finds the set. " +
+            "Use bitgraph_check instead when the user only wants to know whether files are on record.",
         inputSchema: {
             paths: z
                 .array(z.string().min(1))
                 .min(1)
-                .max(MAX_FILES)
-                .describe(`File paths to BitGraph (absolute paths preferred), up to ${MAX_FILES}.`),
+                .max(MAX_PATHS)
+                .describe(`File or directory paths to BitGraph, up to ${MAX_PATHS}; a directory expands to its files, up to ${MAX_MEMBERS} in all.`),
             again: z
                 .boolean()
                 .default(false)
-                .describe("false (default): files already on record are returned as-is, nothing minted. true: make a new fused artifact from every file even if its bytes are already on record. Outcomes are per unique file content: two paths with identical bytes yield one artifact."),
+                .describe("false (default): files already on record are returned as-is, nothing made. true: put every file in the set even if its bytes are already on record. Outcomes are per unique file content: two paths with identical bytes are one member."),
             response_format: responseFormatSchema,
         },
         annotations: {
@@ -136,18 +217,42 @@ export function buildServer(deps = {}) {
             idempotentHint: false,
             openWorldHint: true,
         },
-    }, async ({ paths, again, response_format }) => {
+    }, async ({ paths, again, response_format }, extra) => {
         const config = configFromEnv();
+        const report = progressReporter(extra);
         try {
-            const read = await readPaths(paths);
-            // Unique by content, first path wins for the artifact's name; extra paths listed too.
-            const byDigest = new Map();
-            read.forEach((r, i) => {
-                const entry = byDigest.get(r.digestB64) ?? { bytes: r.bytes, paths: [] };
-                entry.paths.push(paths[i]);
-                byDigest.set(r.digestB64, entry);
+            // 0. Paths to files, before any network call.
+            const expanded = await expandPaths(paths, MAX_MEMBERS);
+            const files = expanded.files;
+            if (files.length === 0) {
+                return fail("Error: nothing to BitGraph: the given directories hold no regular files (hidden entries and symbolic links are left out).");
+            }
+            // 1. Evidence from an earlier set that the site has not indexed yet goes first.
+            if (pendingIndexCount() > 0) {
+                const flushed = await flushIndex(config, report);
+                if (flushed.pending > 0) {
+                    return fail(`Error: ${flushed.pending} members of an earlier set are still waiting to be indexed and the site could not take them; nothing was BitGraphed. Run bitgraph_record again in a moment: the waiting evidence is sent first.`);
+                }
+            }
+            // 2. The scan: one pass per file.
+            let scanned = 0;
+            report(0, files.length, `hashing ${files.length} files`);
+            const scans = await mapConcurrent(files, SCAN_CONCURRENCY, async (p) => {
+                const s = await scanFile(p);
+                scanned += 1;
+                report(scanned, files.length, `hashed ${scanned} of ${files.length}`);
+                return s;
             });
+            // 3. Unique by content; the first path names the member, every path is reported.
+            const byDigest = new Map();
+            for (const s of scans) {
+                const entry = byDigest.get(s.digestB64) ?? { file: s, paths: [] };
+                entry.paths.push(s.path);
+                byDigest.set(s.digestB64, entry);
+            }
             const unique = [...byDigest.keys()];
+            // 4. What is on record already.
+            report(0, 1, "checking the ledger");
             const checked = await batchCheck(config, unique.map(toUrlSafeB64));
             const existing = new Map();
             for (const d of unique) {
@@ -155,92 +260,147 @@ export function buildServer(deps = {}) {
                 if (entry && entry.proofs.length > 0)
                     existing.set(d, entry.proofs);
             }
-            const toMint = again ? unique : unique.filter((d) => !existing.has(d));
-            const fused = new Map();
-            const failed = new Map();
-            for (const d of toMint) {
-                const entry = byDigest.get(d);
-                const name = entry.paths[0].split(/[\\/]/).pop() ?? "file";
+            // 5. The set: every fresh file (every file, with again), one call.
+            const excluded = new Map();
+            const toMint = [];
+            for (const d of again ? unique : unique.filter((x) => !existing.has(x))) {
+                const f = byDigest.get(d).file;
+                if (f.state === null && f.size > MAX_LOADED_BYTES) {
+                    excluded.set(d, "the file changed while it was read and is too large to read again in memory; run bitgraph_record again for it");
+                    continue;
+                }
+                toMint.push(f);
+            }
+            const attempted = new Set(toMint.map((f) => f.digestB64));
+            let made = null;
+            let failure = null;
+            if (toMint.length > 0) {
+                const kind = toMint.length > MAX_SET_MEMBERS ? "set/2" : "set/1";
                 try {
-                    fused.set(d, await fuseFile(entry.bytes, name, config));
+                    made = await fuseSetPipeline(toMint, config, {
+                        set: kind,
+                        onProgress: (p) => report(p.done, p.total, `${PHASES[p.phase]} ${p.done} of ${p.total}`),
+                    });
                 }
                 catch (err) {
-                    failed.set(d, err instanceof Error ? err.message : String(err));
+                    failure = setFailureText(err);
+                }
+            }
+            // 6. A set/2 lands with only its root on the ledger; its members are indexed afterwards.
+            let index = null;
+            if (made !== null && made.set === "set/2") {
+                const counter = made.proof.commit?.counter;
+                const epochId = made.proof.commit?.epochId;
+                const members = made.members.map((m) => m.memberProof).filter((e) => e !== undefined);
+                if (counter !== undefined && epochId !== undefined && members.length > 0) {
+                    pendingIndex.push({ setDigest: toUrlSafeB64(made.artifactDigestB64), epoch: toUrlSafeB64(epochId), counter, members });
+                    index = await flushIndex(config, report);
                 }
             }
+            // 7. Outcomes.
+            let setOutcome = null;
+            const memberOf = new Map();
+            if (made !== null) {
+                for (const m of made.members)
+                    memberOf.set(m.originDigestB64, m);
+                const { counter, epoch } = positionOf(made.proof);
+                setOutcome = {
+                    set: made.set,
+                    count: made.count,
+                    counter,
+                    epoch,
+                    artifact_digest: toUrlSafeB64(made.artifactDigestB64),
+                    proof_url: proofUrl(config.baseUrl, made.artifactDigestB64, counter ?? undefined, made.proof.commit?.epochId),
+                    manifest_echoed: made.manifestEchoed,
+                    recovered: made.recovered,
+                    index,
+                };
+            }
             const outcomes = [];
-            const frames = {};
             for (const [digest, entry] of byDigest) {
-                const made = fused.get(digest);
+                const m = memberOf.get(digest);
                 const prior = existing.get(digest);
                 for (const path of entry.paths) {
-                    if (made) {
-                        const { counter, epoch } = positionOf(made.proof);
-                        frames[toUrlSafeB64(made.artifactDigestB64)] = made.frame;
+                    const base = { path, digest: toUrlSafeB64(digest) };
+                    if (m !== undefined && made !== null && setOutcome !== null) {
                         outcomes.push({
-                            path,
-                            digest: toUrlSafeB64(digest),
+                            ...base,
                             outcome: "fused",
-                            artifact_digest: toUrlSafeB64(made.artifactDigestB64),
-                            placement: made.placement,
-                            counter,
-                            epoch,
+                            artifact_digest: toUrlSafeB64(m.artifactDigestB64),
+                            placement: m.placement,
+                            counter: setOutcome.counter,
+                            epoch: setOutcome.epoch,
+                            member: m.manifestIndex + 1,
+                            member_count: made.count,
                             total_positions: (prior?.length ?? 0) + 1,
-                            proof_url: proofUrl(config.baseUrl, made.artifactDigestB64, counter ?? undefined, made.proof.commit?.epochId),
+                            proof_url: proofUrl(config.baseUrl, digest, setOutcome.counter ?? undefined, made.proof.commit?.epochId),
                         });
                     }
-                    else if (prior && !failed.has(digest)) {
-                        const first = prior[0]?.proof;
-                        const { counter, epoch } = first ? positionOf(first) : { counter: null, epoch: null };
+                    else if (prior && !attempted.has(digest) && !excluded.has(digest)) {
+                        const first = prior[0];
+                        const { counter, epoch } = first ? positionOf(first.proof) : { counter: null, epoch: null };
+                        const row = first?.member;
                         outcomes.push({
-                            path,
-                            digest: toUrlSafeB64(digest),
+                            ...base,
                             outcome: "on record",
                             artifact_digest: null,
                             placement: null,
                             counter,
                             epoch,
+                            member: row ? row.index + 1 : null,
+                            member_count: row ? row.count : null,
                             total_positions: prior.length,
                             proof_url: proofUrl(config.baseUrl, digest),
                         });
                     }
                     else {
                         outcomes.push({
-                            path,
-                            digest: toUrlSafeB64(digest),
+                            ...base,
                             outcome: "not fused",
                             artifact_digest: null,
                             placement: null,
                             counter: null,
                             epoch: null,
+                            member: null,
+                            member_count: null,
                             total_positions: prior?.length ?? 0,
                             proof_url: null,
-                            error: failed.get(digest) ?? "not attempted",
+                            error: excluded.get(digest) ?? failure ?? "not attempted",
                         });
                     }
                 }
             }
+            // Rows that need reading come first, so a cap drops fused rows, which all share one position.
+            const order = { "not fused": 0, "on record": 1, fused: 2 };
+            outcomes.sort((a, b) => order[a.outcome] - order[b.outcome]);
+            const listed = outcomes.slice(0, ROW_CAP);
+            const omitted = outcomes.length - listed.length;
+            const summary = {
+                files: files.length,
+                directories: expanded.directories,
+                fused: outcomes.filter((o) => o.outcome === "fused").length,
+                on_record: outcomes.filter((o) => o.outcome === "on record").length,
+                not_fused: outcomes.filter((o) => o.outcome === "not fused").length,
+            };
             const structured = {
-                results: outcomes,
-                frames,
-                summary: {
-                    fused: outcomes.filter((o) => o.outcome === "fused").length,
-                    on_record: outcomes.filter((o) => o.outcome === "on record").length,
-                    not_fused: outcomes.filter((o) => o.outcome === "not fused").length,
-                },
+                set: setOutcome,
+                results: listed,
+                omitted,
+                summary,
             };
-            if (failed.size > 0) {
-                const guidance = again
-                    ? "A file marked 'not fused' MAY still have been BitGraphed if the failure was a timeout: run bitgraph_check on it first, then re-run bitgraph_record with again=true for only the files still missing."
-                    : "Re-run bitgraph_record with the same paths: files already on record come back as 'on record' and only the missing ones are BitGraphed.";
+            const markdown = renderRecordMarkdown(outcomes, setOutcome, omitted);
+            if (summary.not_fused > 0) {
                 return {
                     isError: true,
-                    content: [{ type: "text", text: `${fused.size} of ${toMint.length} files were BitGraphed; ${failed.size} failed. ${guidance}\n\n${renderRecordMarkdown(outcomes)}` }],
+                    content: [{ type: "text", text: `${failure ?? `${summary.not_fused} file(s) were left out.`}\n\n${markdown}` }],
                     structuredContent: structured,
                 };
             }
-            const text = response_format === "json" ? capJson(structured).text : renderRecordMarkdown(outcomes);
-            return ok(text, structured);
+            if (response_format === "json") {
+                const full = { ...structured, set: setOutcome !== null && made !== null ? { ...setOutcome, proof: made.proof } : null };
+                return ok(capJson(full).text, structured);
+            }
+            return ok(markdown, structured);
         }
         catch (err) {
             return fail(errorText(err));
@@ -249,18 +409,18 @@ export function buildServer(deps = {}) {
     server.registerTool("bitgraph_check", {
         title: "Check for BitGraphs",
         description: "Check whether files or digests are on record in the BitGraph ledger, without recording anything. " +
-            "Accepts file paths (hashed locally; only digests are sent) and/or raw SHA-256 digests in standard or URL-safe base64. " +
-            "Returns, per item: on_record (the bytes are on record, as an exact recording or as the original a new file was made from), every position by counter, and the proof page URL. " +
+            `Accepts file paths and directory paths (every regular file under them, up to ${MAX_CHECK_FILES} in all; hashed locally, only digests are sent) and/or raw SHA-256 digests in standard or URL-safe base64. ` +
+            "Returns, per item: on_record (the bytes are on record, as an exact recording, as the original a new file was made from, or as a member of a set), every position by counter, and the proof page URL. " +
             "Read-only. Use bitgraph_record to BitGraph files that turn out not to be on record.",
         inputSchema: {
             paths: z
                 .array(z.string().min(1))
-                .max(MAX_FILES)
+                .max(MAX_PATHS)
                 .optional()
-                .describe("File paths to check."),
+                .describe("File or directory paths to check."),
             digests: z
                 .array(z.string().min(1).max(100))
-                .max(MAX_FILES)
+                .max(MAX_CHECK_FILES)
                 .optional()
                 .describe("SHA-256 digests, base64 (standard or URL-safe form)."),
             response_format: responseFormatSchema,
@@ -276,8 +436,9 @@ export function buildServer(deps = {}) {
         try {
             const inputs = [];
             if (paths && paths.length > 0) {
-                const hashed = await hashPaths(paths);
-                hashed.forEach((d, i) => inputs.push({ label: paths[i], standardDigest: d }));
+                const { files } = await expandPaths(paths, MAX_CHECK_FILES);
+                const hashed = await hashPaths(files);
+                hashed.forEach((d, i) => inputs.push({ label: files[i], standardDigest: d }));
             }
             for (const d of digests ?? []) {
                 const trimmed = d.trim();
@@ -289,11 +450,14 @@ export function buildServer(deps = {}) {
             if (inputs.length === 0) {
                 return fail("Error: provide at least one of paths or digests.");
             }
+            if (inputs.length > MAX_CHECK_FILES) {
+                return fail(`Error: ${inputs.length} items; check at most ${MAX_CHECK_FILES} at a time.`);
+            }
             const checked = await batchCheck(config, inputs.map((i) => toUrlSafeB64(i.standardDigest)));
             const outcomes = inputs.map((input) => {
                 const entry = checked.results[toUrlSafeB64(input.standardDigest)];
                 const proofs = entry?.proofs ?? [];
-                const positions = proofs.map((p) => positionOf(p.proof));
+                const positions = proofs.map((p) => ({ ...positionOf(p.proof), ...(p.member ? { member: p.member } : {}) }));
                 return {
                     input: input.label,
                     digest: toUrlSafeB64(input.standardDigest),
@@ -320,7 +484,7 @@ export function buildServer(deps = {}) {
     });
     server.registerTool("bitgraph_get_proof", {
         title: "Get a BitGraph proof",
-        description: "Fetch a BitGraph proof and its context: causal position, every position the same bytes occupy, and the two-sided Ethereum anchor window " +
+        description: "Fetch a BitGraph proof and its context: causal position, every position the same bytes occupy, the row a set member holds (one of N), and the two-sided Ethereum anchor window " +
             "('BitGraphed between X and Y'). Look up by digest (base64, either form), by BitGraph number (e.g. '4523' or '#4,523', current epoch), or by file path (hashed locally). " +
             "Exactly one of digest, number, or path is required. Read-only. " +
             "markdown returns a summary; json returns the full proof object with positions and anchor window.",
@@ -387,7 +551,7 @@ export function buildServer(deps = {}) {
             const selEpoch = epoch !== undefined ? toUrlSafeB64(fromUrlSafeB64(epoch)) : undefined;
             const detail = await getProofDetail(config, urlSafeDigest, selCounter, selEpoch);
             if (detail.proofs.length === 0) {
-                return fail(`Not on record: no proof exists for digest ${urlSafeDigest}. Use bitgraph_record to record the file.`);
+                return fail(`Not on record: no proof exists for digest ${urlSafeDigest}. Use bitgraph_record to make a BitGraph of the file.`);
             }
             if (response_format === "json") {
                 const capped = capJson(detail);