Omakase CPHEEO
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
omakase-cpheeo
Search and cite the CPHEEO manuals (Government of India) over MCP and REST. Every result carries the manual, edition, part, clause or table, and printed page, with Omakase attribution.
Indexed now: Manual on Sewerage and Sewage Treatment Systems, Part A: Engineering (2013), including all 51 source-listed appendices/sheets, and Manual on Water Supply and Treatment Systems, Parts A, B and C (2024). Storm water (2019) and solid waste (2016) are next. See docs/plan.md.
Requirements
Python 3.13 and uv
The source PDFs in a local folder (not in this repo). Default
~/Downloads/CPHEEO; override withOMAKASE_CPHEEO_SOURCES.
Related MCP server: Lookout
Use
uv run omakase-cpheeo ingest # parse the manuals, build data/build/cpheeo.sqlite (all four source manuals)
uv run omakase-cpheeo search minimum slope for 200 mm sewer
uv run python scripts/visual_coverage.py # page review and transcription coverage
uv run python scripts/check_verified_retrieval.py # numerical/formula retrieval and provenance
uv run omakase-cpheeo clause SEW-2013-A:T3.2
uv run omakase-cpheeo eval -v # score retrieval against eval/questions.yaml
uv run omakase-cpheeo eval --questions eval/search-ranking.yaml -k 1 -v # diameter ranking regressions
uv run omakase-cpheeo serve --port 8765 # MCP at /mcp, REST at /v1/*
uv run omakase-cpheeo eval --questions eval/sewage-appendices.yaml -v # appendix retrieval
uv run omakase-cpheeo eval --questions eval/paraphrase-groups.yaml -v # rephrasings of one question, scored on the worst
uv run python scripts/check_appendix_coverage.py # compare all appendix pages with source contents
uv run python scripts/mcp_smoke.py # exercise every MCP tool against a running serverInterface
MCP tools (streamable HTTP at /mcp): search, fetch, get_clause, get_toc, list_manuals. search and fetch follow the result and document shape ChatGPT connectors use.
Coverage. Every search response has
coverage: the manuals searched and astatus(ok,no_match,requested_manual_not_indexed,topic_may_be_outside_corpus,domain_not_indexed,manual_not_indexed,unknown_domain,unknown_manual). Manuals known but not yet indexed (storm water 2019, solid waste 2016, sewerage Parts B/C) are listed incorpus.UNINDEXEDand returned bylist_manualsundernot_indexed.Evidence precedence.
textis the extracted PDF text layer, never edited. Verified transcriptions take precedence for what they cover.text_qualitylists known extraction defects (curated indata/text-defects.json, each tied to its verified transcription) and possible omissions (skipped items in "Firstly … Thirdly" or "(a) … (d)" lists).reading_textistextwith both marked inline.fetchlabels each layer. Search results carryquality_flags.Errors. Unknown ids, manuals and kinds return
{"error": {"code", "message", "suggestions", "recovery"}}(MCPisErrorresults; REST 404/400).
REST:
Endpoint | Returns |
| Ranked units with citation and snippet |
| Section text, native table rows and source-image references |
| Manuals and unit counts |
| Chapters, sections and tables |
| Readable clause page with Omakase attribution (what result |
| Source-page image for a table, figure or scanned content |
| Liveness and loaded manuals |
Unit ids are citation handles: SEW-2013-A:3.6 (section), SEW-2013-A:T3.2 (table), SEW-2013-A:F3.4 (figure), WS-2024-A:AN6.1 (annexure), WS-2024-A:AN6.3-T4 (Table 4 inside Annexure 6.3; annexures number their tables from 1).
Sewage Parts B (Operation and Maintenance) and C (Management) are deferred until their published source PDFs are available locally.
The sewage appendices have separate handles such as SEW-2013-A:A3.2B (lettered sheet), SEW-2013-A:TA7.3-1 (appendix table) and SEW-2013-A:FA3.2A-1 (appendix figure). The old grouped handles (A3.2, A3.4, A3.5) still return both sheets. Appendix titles and page boundaries come from the manual's native-text contents, and all 185 appendix pages are available as full-page images. Lookup accepts printed references such as "Appendix A 3.2 B" or "Table A 7.3- 1".
Scanned tables and figures retain source imagery. Visually transcribed and separately rechecked text is available where explicitly marked transcription_status: verified: cells under verified_transcriptions[].rows, searchable text under verified_text, and LaTeX/readable representations for equations. Original images remain available alongside exact source-region crops, PDF hashes and verification observations. Equation handles include SEW-2013-A:E5.30 and also appear on their parent sections.
Visual transcription proceeds in small targeted batches. There are 295 verified records (136 tables, 159 equations) and one unclear formula. Select a few known tables/equations from the inventory or retrieval needs, inspect their source crops and necessary continuations, and verify those records. An exhaustive page sweep is not required. The historical page ledger is audit metadata, not a queue of thousands of pages to inspect. See visual transcription coverage and workflow. No OCR is run. Native water table text is not automatically visually verified; unclear math retains its source image and explicit note.
Verified values, table headers, notes and mathematical text are searchable. get_clause and fetch include the text, provenance and MCP image blocks; REST returns the same data and image references; clause pages display text and source images. Verify the returned record's scope and any source-error notes before interpreting numerical values.
Settings
Variable | Default | Purpose |
|
| Folder holding the source PDFs |
|
| Index used by the server |
| unset | Link included in every result's attribution |
| unset | Public base URL (tunnel or deployment); adds result links and allows that Host |
| unset | Extra comma-separated browser origins to accept |
|
| PNG resolution for scanned tables and other embedded pictures |
|
| JPEG resolution for full figure pages |
Branding
Assistants often skip server instructions, so the Omakase brand travels in the data they copy:
Every result's
citationends with· via Omakase CPHEEO.citation_plainis the same reference without the suffix, for pasting into reports.Every result carries
source_name: "Omakase CPHEEO", andsource_icon(the icon URL) whenOMAKASE_CPHEEO_PUBLIC_URLis set.fetchtitles start withOmakase CPHEEO ·.fetchtext andget_clause(source_note, its first field) carry one line:Source: Omakase CPHEEO search (free CPHEEO lookup by Omakase), plus the clause page URL when a public URL is set.The MCP server is named "Omakase CPHEEO" (title "Omakase CPHEEO: search and cite CPHEEO manuals") and declares the Omakase icon:
<public URL>/icon.png, or an inlinedata:PNG without a public URL. Tool titles start with "Omakase CPHEEO:"./icon.png(256 px),/icon-1024.pngand/favicon.icoserve the icon (assets/, from the Omakase app icon). Clause pages show it next to "Omakase CPHEEO".
When adding the connector in ChatGPT or Claude, name it "Omakase CPHEEO", and upload assets/omakase-icon.png as its icon if the form allows one.
License
Copyright (C) 2026 Ajinkya Puranik. The project's original code and documentation are licensed under the GNU Affero General Public License, version 3 only (AGPL-3.0-only).
CPHEEO manuals, excerpts, transcriptions of manual content and source imagery retain their original copyrights and are not relicensed under the software license. The manuals are Government of India copyright and may not be reproduced for commercial purposes without MoHUA permission. The free service returns cited excerpts; it does not redistribute the PDFs.
Test from ChatGPT or Claude (temporary public URL)
mise exec cloudflared@latest -- cloudflared tunnel --url http://127.0.0.1:8765 # prints https://<random>.trycloudflare.com
OMAKASE_CPHEEO_PUBLIC_URL=https://<random>.trycloudflare.com uv run omakase-cpheeo serveAdd https://<random>.trycloudflare.com/mcp as a custom MCP connector with no authentication. The URL changes every time the tunnel restarts.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Connectors
Hosted MCP server for finding authoritative primary data sources and official portals.
MCP server for FCA Handbook compliance analysis with verbatim citations.
Unified 15-tool SEO, SERP, clean Markdown & GEO (LLM citation Share-of-Voice) MCP server.
The Water Hub: the site's own MCP server — dataset; every answer cites the site.
Related MCP Servers
- FlicenseAqualityCmaintenanceMCP server for querying a page-citable research knowledge base built from PDFs, with exact filename and page citations.6-
- AlicenseNot gradedqualityFmaintenanceMCP server enabling local-first web search, fetch, extract, and caching with citeable excerpts, no API key required. Supports research workflows for agents and apps.96 npm1MIT
- AlicenseAqualityDmaintenanceThis server crawls documentation sites and provides search and page-retrieval tools through MCP, enabling users to query indexed documentation content.4MIT
- FlicenseAqualityCmaintenanceLocal-first, read-only MCP server for retrieving and searching authorised CCSDS and ECSS standards, enabling citation-rich passage-level search and document access.3-