Skip to content

eDiscovery Decoder MCP

In document review, the number that decides whether a production holds up is not generated by a model, and a model that estimates it is a liability. This server is the other kind of tool: deterministic calculators for the statistics a Technology-Assisted Review workflow has to state, wrapped in a free keyless endpoint, with the arithmetic declared in the tool descriptions.

Server type: Remote (Streamable HTTP)
Endpoint:    https://mcp.ediscoverydecoder.com/mcp
Auth:        none (free educational, non-commercial preview)
Tools:       15
Registry:    com.ediscoverydecoder/mcp
Price:       free

What the endpoint says about itself

A keyless initialize on https://mcp.ediscoverydecoder.com/mcp answers 200 as ediscovery-decoder-news-calc 0.1.0 on protocol 2025-06-18, and the instructions are unusually explicit about what the server is and is not:

eDiscovery Decoder (free educational non-commercial preview): curated eDiscovery and legal-tech news plus deterministic Technology-Assisted Review (TAR) and document-review statistics calculators.

tools/list returns 15 tools in three groups. The calculators are the substance:

Tool What it computes
calculate_review_metrics Score a coded sample from a full confusion matrix (true and false positives and negatives), for example comparing two reviewers or two model generations
calculate_elusion Estimate how much responsive material remains in the set that was not reviewed
calculate_sample_size How many documents to sample randomly to estimate a proportion such as richness or elusion at a target confidence
calculate_tar_recall_estimate Overall TAR recall and how many responsive documents were missed, combining the count already found with the sample
calculate_prevalence_richness The share of a population that is responsive or relevant, from the positive counts
calculate_control_set_recall Recall against a known control set: the share of documents already confirmed relevant that the workflow found
compare_tar_cutoffs Candidate score or rank cutoffs side by side: how many documents sit above each, and the trade each implies
validate_sample_design QC a validation sampling plan against the documented elements a defensibility declaration needs

The rest is orientation and reference: ping, list_capabilities, get_demo_guide, get_resource_content for the JSON behind an edd:// resource (demo guide, TAR learning path, glossary, news), get_prompt_template for guided prompts such as a matter kickoff or a weekly digest, and two news tools, search_news and get_news_brief, over a curated eDiscovery and legal-AI feed.

Why deterministic arithmetic earns a catalogue entry

An agent with no calculator will happily invent a recall figure, and the failure is invisible because the number looks plausible. Here the tool either computes a defined quantity from inputs the reviewer supplies or refuses on missing inputs, and validate_sample_design goes one step further by checking a plan for the elements a defensibility narrative is expected to contain rather than producing a number at all. The client-visible claim is narrow and that is the point: it does the math, the reviewer owns the methodology.

It also costs a reviewer nothing to check. The endpoint is keyless, so a model can be told "use the calculator for that figure" without an account, a card, or a procurement step, and list_capabilities plus get_demo_guide give it a scripted path for the first call.

Limits worth writing down

  • Educational, non-commercial preview. The vendor says so on both the submission and the MCP page: for educational use, not legal advice, outputs to be reviewed by a qualified human. Treat it as a calculator a professional uses, not a service an operator delegates to.
  • Aggregate counts only, no case files. Nothing here ingests a production set; the reviewer supplies the counts.
  • No public repository, so the formulas cannot be read in source, only in the tool descriptions and the returned workings.
  • Two of the fifteen tools are news, so the usable calculator surface is thirteen.
  • No price and therefore no commercial commitment. A free preview can change terms; nothing was promised about uptime.

Where it sits in this catalog

The catalog's legal coverage is legal research and court-record retrieval (LEGAION and the regional legal entries). Review statistics are a different job: not what the law says, but whether a document review can be defended with the numbers it reports. That is a step most operators only meet once, usually under a deadline, and it is the exact place a chat model should be handed a calculator instead of a prompt.

Frequently asked questions

What does eDiscovery Decoder MCP cost?

Nothing. It is a free keyless endpoint with no account and no API key, published by the vendor as an educational, non-commercial preview.

What can the calculators actually compute?

Recall and precision from a confusion matrix, elusion, the sample size needed for a target confidence, a TAR recall estimate, prevalence and richness, recall against a control set, a side-by-side comparison of score or rank cutoffs, and a QC pass over a validation sampling plan.

Does it read case files?

No. It takes aggregate counts the reviewer supplies and returns the arithmetic; no production set is ingested and no document is reviewed by the server.

No. The vendor states on both the submission and the MCP page that it is educational, not legal advice, and that outputs should be independently reviewed by a qualified human.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing