MCP Server Discovery - November 2, 2026 (Night Sweep)¶
Summary¶
Six hours after the evening pass. The mcp.so recentServers block carried 30 cards with a date and the newest at 2026-11-02T09:04:11Z, so exactly two sat above the evening stamp of 2026-11-02T01:54:34Z: docuwriter-ai and signal8. The mcpservers.org /all page behaved as the two previous passes recorded and the counter moved by two, from 18,883 to 18,885, with catalogian-mcp and opsis-osint/opsis-mcp at the top.
Two arrivals earned guides. Catalogian MCP turns a CSV or JSONL source into a versioned record with its own MCP endpoint and a delta per ingest, and it was driven end to end keyless, including a revision that returned 100 new and 100 deleted rows. Signal8 MCP computes what the cap table does next: a dilution pressure score, baby shelf capacity against the live float and a cash runway calendar, over 109 measured tools, and no other guide in this catalogue mentions dilution at all.
Sources Swept¶
| Source | Read | Result |
|---|---|---|
mcp.so homepage recentServers |
direct, 30 cards with a date | newest 2026-11-02T09:04:11Z, two above the evening stamp 2026-11-02T01:54:34Z |
mcpservers.org /all |
r.jina.ai, X-Return-Format: html, read twice |
first read 18,883 (the evening pass counter, stale), second read 18,885 live, 30 slugs |
| Endpoint probes | keyless JSON-RPC | 2 endpoints open on discovery, 2 OAuth 401, 1 open with a working anonymous session |
The two read rule held for a third pass. The first /all fetch returned the same counter and the same thirty slugs the evening pass had dispositioned, and the second fetch, cache busted with a nanosecond stamp and a different user agent, returned 18,885 with the page turned over. The read whose counter has moved is the live one.
New Servers Catalogued¶
Catalogian MCP - A Data Record Your Agent Can Question¶
Relevance: HIGH. Endpoint driven end to end (initialize HTTP 200, serverInfo catalogian 1.1.0, 18 tools listed, and live calls across schema, query, filter, search, editions, delta, edition diff and export).
The product is a record: a CSV, TSV or JSONL source from a URL or an upload, keyed on a column, indexed, and versioned on every ingest. Each record ships its own MCP endpoint and eighteen read only tools. Onboarding needs no account: POST /v1/ephemeral/ingest with a URL returns a session token good for twenty four hours, three ingests and fifty tool calls, with 10,000 rows indexed.
Measured on a public dataset: 300 rows and 117 fields ingested under key field id; snapshot_schema returned the types, null rates and example values; query_snapshot with group_by on permit_type returned the full distribution, 299 and 1, in one object; filter_snapshot_rows returned the single matching row; search_snapshot returned rows ranked by relevance. A second ingest at a different slice landed on the same record and produced an exact delta of 100 new, 100 deleted and 200 unchanged, which list_editions recorded as edition 2 with kind: revision and get_edition_diff returned at row level with field level values.
It earns a guide because nothing in this catalogue treats a file as a versioned object. Settra MCP syncs sheets and files into PostgreSQL behind a semantic layer and answers what an approved definition says; this answers which rows arrived this week and what value moved, which is a different question about a single file.
Guide: Catalogian MCP
Signal8 MCP - Dilution Intelligence for US Filings¶
Relevance: HIGH. Discovery verified (keyless initialize HTTP 200, serverInfo signal8-mcp 0.19.0, tools/list 109 tools in 221,131 bytes), calls gated (seven tools attempted keyless, all JSON-RPC -32001 Authentication required).
The slice is what the cap table does next. get_dilution_coverage is a cheap precheck, get_dilution_risk returns a Dilution Pressure Score from 0 to 100 across seven components, get_baby_shelf_capacity recomputes the Form S-3 General Instruction I.B.6 one third of float cap against the live float, and get_dilution_instruments, get_dilution_snapshot and get_dilution_history complete the family. Cash runway, must raise screening, filing text and structure, insider and institutional ownership, and a congressional trading family with profit and loss and FEC donors run beside it.
It earns a guide on a catalogue gap rather than on novelty alone: a grep across all 1,082 guides returns no entry mentioning dilution, baby shelf or cash runway, and the nine EDGAR entries here read what a company reported rather than what can still arrive.
Guide: Signal8 MCP
Also Identified (Not Catalogued)¶
From the mcp.so block, the newest card above the evening stamp besides Signal8 is docuwriter-ai (DocuWriter.ai, Developer Tools, 401 with OAuth protected resource metadata on https://app.docuwriter.ai/mcp, 77 tools documented over codebase documentation generation, held as developer documentation tooling rather than an operator data surface).
From /all, the fresh slugs are all vendors of covered classes: catalogian-mcp is the new entry; opsis-osint/opsis-mcp is Opsis OSINT, the agent OSINT surface held at the evening pass on the OpenOSINT MCP precedent, OAuth with three free searches and no new disposition; testedmcp-com answers a keyless initialize on https://testedmcp.com/mcp as testedmcp 0.1.0 with read only lookups over 197 installed and handshake tested MCP servers, held as a directory and benchmark fixture, which is this catalogue's own subject; nexusbro-com is the NexusBro surface held on October 31 against eight SEO audit entries; heritagedeed-com-developers is the entry catalogued at the evening pass; shipvo/shipvo is the Shipvo developer mobile testing hold from October 31; alfa-reza/cvss-calculator-mcp and the remaining personal handle slugs are listings without a product surface; nova-civai-co-agent-content-pipeline-agent scans trends, generates drafts and publishes for one vendor's own products behind a Nova account, held as single vendor product tooling; and abullycoin-com serves facts about one meme token on Robinhood Chain, held as consumer crypto.
The docuwriter-ai card, testedmcp-com and nova-civai-co-agent-content-pipeline-agent were read in full before the hold. The submission window on chatmcp/mcpso was outside this run's scope and was not read.
Core Listing Verification¶
| Surface | Result |
|---|---|
mcpservers.org /servers/corpusiq-io |
not re-probed this run |
| Glama and the official registry | not re-probed this run |
| mcp.so four direct slugs | not re-probed this run |
The listing probes are recorded as not run rather than presumed green, and the next pass should re-probe all five surfaces.
Verification¶
- Both guides were written against measured responses, not vendor pages. Catalogian was driven across nine tools with a real dataset, a real revision and real deltas. Signal8 was measured at discovery and gated at call time, and the gate is stated in the guide rather than smoothed.
- Two defects are recorded rather than resolved.
download_snapshoton Catalogian answers an S3 signature mismatch, so the export path is broken while the query path works.compare_snapshotsreported Removed 0 on an edge whereget_deltareporteddeletedCount: 100, so deletions should be read throughget_deltaorget_edition_diff. - A vendor sample file exposed a byte order mark defect: the key field was taken as the BOM prefixed
IDand the revision was then refused withKEY_FIELD_MISSINGon headers that visibly contain the column. - Guide tree counted from disk: 1,082 server directories before, 1,084 after, 1,085 index files under the guide tree.