DondeGo Entity Resolver

DondeGo Entity Resolver#

Restoration release, 2026-10-16. Bounded requests for previously unseen entities are supported. The seven exact request/result catalogue and recursive adult-category closure have been removed. Identity uncertainty, contradictions, shared quotas and Wikimedia coordination remain enforced. See current controls and measured verification.

Compare a public place, person, organization or work with Wikidata records and understand what supports a match. Search for candidates, inspect selected facts, check an existing identifier, and keep ambiguity visible when evidence is missing.

Independent developer and publisher: Aleksandr Prokofyev. The publishing identity is an individual, not a claim of business verification. DondeGo Entity Resolver is the application name. The public source repository is maintained under the same developer's GitLab account.

This is an independent application. It is not made, sponsored or endorsed by the Wikimedia Foundation, Google or OpenAI. It does not use their logos as its product identity. Wikidata is the source of public structured data, not the publisher.

A workflow beyond a search result#

The application runs its own deterministic Python resolver before returning a decision. The public remote service uses the same core as the local package:

  • Audit a supplied identifier: inspect that item's facts first and explain supporting evidence or contradictions; do not silently replace the supplied ID.
  • Compare candidates: evaluate known, public entity facts against candidates using kind-specific rules and explicit evidence codes.
  • Keep missing evidence separate from conflict: an absent statement is not proof that two records describe different entities.
  • Preserve uncertainty: a matching name alone does not approve an identity. HOLD, AMBIGUOUS and INSUFFICIENT_EVIDENCE are useful results.
  • Return compact evidence: selected facts and bounded relationships make the result inspectable. A confidence-like provider score is not identity proof.

The identity workflow, resolver rules and reason-code reference describe these operations. The code is public: identity implementation, resolver and offline tests. The documented benchmark is not an independent measurement of identity accuracy. No accuracy guarantee is claimed.

Try a bounded task#

  • “Find candidates for Fox Theatre and explain why the name alone is ambiguous.”
  • “Check whether Q76 is the right Wikidata record for the Fox Theatre.”
  • “Read selected facts for Q1440190 and explain what they establish.”

Use only facts about the public entity that are relevant to your request. This is public-record reconciliation, not identity verification of a person, KYC, background screening or a people-tracking service. Do not send private personal records, user GPS location, home addresses, credentials, health or payment data, or government-issued identifiers. QIDs identify public Wikidata records.

The application never modifies Wikidata, your files or your databases, and never publishes a sameAs link. A returned identity candidate still needs your own review before downstream use. No local installation is needed for the remote tools, and no sign-in or provider key is requested.

Remote service and data source#

Endpoint: https://wikidata-knowledge-mcp.revanalex.workers.dev/mcp

The public remote service uses Wikidata only. Google Knowledge Graph is not available remotely. The underlying local PyPI/stdio package retains its existing name, wikidata-google-knowledge-mcp version 0.3.0, and optional Google corroboration; that local feature is not part of this hosted application.

Wikidata's structured-data licensing permits reuse under CC0. Access uses its documented public services and bounded queries; the application does not bypass provider limits. CC0 does not imply trademark rights, partnership or endorsement. API availability and applicable service terms still apply. See Wikidata data access.

The shared public ceiling is 100,000 HTTP admissions per UTC day, including protocol setup and invalid requests. Exhaustion returns HTTP 429 until midnight UTC. Accounting failure stops execution. Availability is best effort, without an SLA. Each tool has strict input, result and provider-attempt bounds; no public corpus batches, unrestricted graph traversal or arbitrary URL fetching are offered.

Requests are processed to perform the tools. Relevant queries reach Wikimedia services; Cloudflare handles network traffic. The application retains operational aggregates for 31 UTC day buckets rather than full prompts or identity payloads. Read the privacy notice, service terms and remote setup and limits for exact scope and provider handling.

Product support · Confidential security reporting

Demonstration and review evidence#

The 2026-10-16 remote recording shows actual bounded HTTP MCP tool results and identity decisions, with staging/production surfaces and remaining checks labelled. It is not a ChatGPT model-selection demonstration. The distribution verification distinguishes native client discovery, model execution and directory publication.

The OpenAI listing is undergoing remediation after review feedback. This page does not claim that the application is approved or published in that directory.

Documentation for version 0.3.0 · Changelog · Issues