We added some helper commands to the Makefile to
facilitate the development of the docs-mcp.
These commands allow developers to easily
connect to the docs-mcp server with your
favorite LLM.
You can start the MCP Inspector with
`make mcp-inspector` to connect to the docs-mcp server and
inspect the different endpoints and their responses.
Add `src/frontend/servers/mcp`, a stateless TypeScript (Express +
`@modelcontextprotocol/sdk`) MCP server exposing `search_documents`,
`read_document` and `create_document` over Streamable HTTP.
It is an OAuth resource server holding no credentials of its own: it
verifies the caller's Keycloak token against the realm JWKS
(`src/auth/jwtVerifier.ts`), checks the `docs-mcp` audience, then
forwards the same token unchanged to Django's
`/api/v1.0/mcp/documents/*` endpoints (`src/docsApiClient.ts`),
which stays the sole authority on document permissions.
No token exchange.
Ship the `mcp-development` compose service, its `env.d/development/mcp`
env file, a `build-mcp` Makefile helper, and
`documentation/mcp_server.md` describing the full flow and
how to connect MCP clients.
The nb_accesses cache TTL was very short, 30 seconds. That mean that the
user will hit the cache for a very short period and the cache is
probably not be hit. This is what we can see in the slow queries from
the pg_stat_statements table. The query to compute the nb_accesses is
executed a little bit less than the number of queries to list or
retrieve documents, meaning the cache is not used.
We want to configure other options on the redis cache. By default there
is no timeout on the connection to socket and no timeout for read/write
operations. We set default values in all caches used in production. The
settings IGNORE_EXCEPTIONS differ between the default and the session
cache. Activating it behaves like a missed cache. Enabling it for the
session should lead to unwanted side effects, by returning falsy on the
session creation, a retry mechanism of 10000 attempts is made in the
SessionStore.create method, the request can stay in this loop for a long
time.
Release 3.17.2 of DRF now takes care of DATA_UPLOAD_MAX_MEMORY_SIZE
and is checked when the body request is parsed. Before that, DRF wasn't
using it at all and we were only looking for custom settings linked to
the media and conversion file upload. We must now also configure this
setting.
whitenoise middleware is failing a lot with a cancelled exception from
asyncio. Using whitenoise is not needed in our case, we are just serving
an API with django and DRF. We decided to completely remove it.
We have performance issue with the media_auth endpoint. We need to
monitor and profile this endpoint to understand what is happening. We
decided to install and configure django-silk. By default django-silk is
not enabled and not accessible, once enabled migration must be run and
the /silk/ endpoint will be accessible only using a super user.
To have relecant profiles and monitor we must have a database comparable
to a production one because we are facing a scaling issue. Unless a
million documents probably nothing is happening, for this we added
several management commands to help us. A first one to anonymize a
database than can be reused without any risk to leak PII data, can be
called using python manage.py anonymize_database with some mandatory
options. An other command to generate a database profile, this command
should be run on a production database, this profile can then be used by
the generate_volumetry to populate a development database with same
volumetry than describe in the profile.
We have renamed the Chinese locale from `cn_CN`
to `zh_CN` to align with standard locale naming
conventions.
Additionally, we have added support for two new
locales: Esperanto (`eo_PL`) and
Traditional Chinese (`zh_TW`).
We write a documentation explaining how to enable the safe guard
collaboration feature to prevent a user not connected to the websocket
to erase the data coming from users connected to the websocket.
The settings COLLABORATION_WS_NOT_CONNECTED_READY_ONLY contains a typo.
We don't want to see READY_ONLY but READ_ONLY. This commit fix the typo
and change everywhere in the codebase the settings name.
The S3 usage made by docs can be complex. We want to write a
documentation to explain the auth_request nginx module usage and list
known issues with the configuration.
Anthony LC