GeoLens MCP Server

Read-only access to a self-hosted GeoLens spatial catalog: datasets, features, maps, sandboxed SQL.

Publisher claimed. No tool list reported, and Pod has not connected to this server.

Status

Pod has not dialled GeoLens yet, so everything on this page is what its publisher reported rather than what we observed. Registries describe servers; they do not connect to them. Until a check runs, treat the tool list below as a claim.

Connect

Published as geolens-mcp on pypi. Runs locally.

Known issues

12 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded — those are release checklists and internal refactors, not things that will go wrong for you. Showing 9.

Most discussed

Client-side state survives an identity change: drawing-store and search-store are unguarded

Found while fixing a review finding on #1708: the URL-import session was module-scoped state keyed only by URL, so a logout followed by a login without a page reload left it alive and attached the new identity to the previous user's in-flight import and job id.

While fixing it I found stores/analysis-form-store.ts already carrying a comment about this exact bug class, from #793. Someone hit it, fixed that one instance, and the pattern didn't propagate. A sweep of the client-side stores an

Read the thread · 2026-08-30 · open · outside contributor · 0 comments

OpenAPI: declare the cookie-mode auth headers as operation parameters

Found during the docs API-reference sync (getgeolens.com#94, codex review): the cookie-mode browser auth flow shipped in v1.13.0 (#1446) negotiates with X-GeoLens-Auth-Mode: cookie and enforces double-submit CSRF via X-CSRF-Token (backend/app/modules/auth/cookies.py:29-30), but the OpenAPI spec mentions both only in prose. At v1.13.1, /auth/login/, /auth/refresh/, and /auth/logout/ declare zero operation parameters, so generated SDKs and the docs reference's request UI cannot express

Read the thread · 2026-08-15 · closed · outside contributor · 0 comments

sec(catalog): decide one provenance projection — dataset reads, /versions/, and MCP disagree

Three surfaces currently project dataset provenance three different ways, and the strictest one contradicts the loosest one. A decision is needed before #1285 builds Source-panel UI on top of the loosest surface — changing a projection after UI ships on it is a breaking change; deciding now is cheap.

The three disagreeing surfaces (verified on `main` at 403027cba)

GET /datasets/{id} and list reads — DatasetResponse (backend/app/modules/catalog/datasets/domain/schemas.py:264, fi

Read the thread · 2026-08-09 · closed · outside contributor · 1 comment

feat(catalog): project source health and freshness timestamps into search summaries

Search result summaries omit source_health, last_checked_at, and last_refreshed_at, so the search-card health chips added in #1279 stay silent on the search surface and MCP/SDK search reads return null for those fields (documented in #1278). Dataset detail responses already carry all three.

Scope:

Add the three fields to the search summary projection (respecting the existing visibility filtering — the projection must not leak health detail for datasets the caller cannot access).
Regen

Read the thread · 2026-08-08 · closed · outside contributor · 0 comments

feat(mcp): expose origin, freshness, and health in MCP and SDK reads

Blocked by the source-state fields and staleness issues in this milestone.

An agent deciding whether to trust a dataset needs exactly the fields this milestone adds: origin, freshness, health, last refreshed. mcp/geolens_mcp is hand-maintained (like the CLI) and read-only; the SDKs regenerate from openapi.json and pick the fields up mechanically. No competing catalog product ships an MCP server at all, so keeping ours current with the trust metadata is cheap differentiation.

Scope

Read the thread · 2026-08-05 · closed · outside contributor · 1 comment

Most recent

url_redaction: has_url_credentials raises ValueError on a malformed authority, unguarded at the connector-config gate

Problem

has_url_credentials (backend/app/core/url_redaction.py:66) calls urlsplit unguarded, so it has the same failure mode #1119 fixed in its sibling redact_url_credentials:

>>> has_url_credentials("https://.[::1]")
ValueError: Invalid IPv6 URL
>>> has_url_credentials("https://[::1")
ValueError: Invalid IPv6 URL
>>> has_url_credentials("https://user:pw@.[::1]/x")
ValueError: Invalid IPv6 URL

All three reproduce on current main (0506a5260).

Why most call sites are fi

Read the thread · 2026-08-02 · closed · outside contributor · 0 comments

test(mcp): server.py has no tests and CI's path filter excludes backend changes — contract drift invisible

From the 1.6.0 pre-tag audit. The MCP server's tests are mock-only (mcp/), server.py itself is untested, and the CI job's path filter doesn't include backend/** — so a backend API change that breaks the MCP server's read contract runs zero MCP tests.

Fix: a small live-contract test tier (spin the API, exercise each MCP tool against it) and add backend/openapi.json (or backend/**) to the MCP job's path filter.

Read the thread · 2026-07-28 · closed · outside contributor · 0 comments

sec(auth): API-key hardening — no expiry, no staleness check, query-param lane writes the credential into logs

From the 1.6.0 pre-tag audit (latent, P3). Three related gaps in the API-key lane:

No expires_at column: keys live forever unless manually revoked (auth/models.py:176-202).
No token_version staleness check: a role demotion doesn't invalidate previously minted keys the way it does JWTs.
The query-param fallback (_resolve_api_key(), auth/dependencies.py:62-68) puts the credential in request URLs, so it lands in access logs and any upstream proxy logs.

Suggested shape: add optiona