Alexandria MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Alexandria MCPsearch the knowledge base for how to set up OAuth"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Alexandria
Alexandria serves a versioned Markdown/text knowledge base through MCP. The Knowledge repository is the source of truth; this repository contains the runtime, index preparation, tests, image definitions, deployment contracts, and operating tools.
From a GitHub knowledge repository to a working MCP
flowchart LR
G[Private GitHub Knowledge repo] --> C[Checkout exact commit]
C --> I[Build LanceDB index]
I --> B[Verified data block]
A[Alexandria code] --> D[Code-only image]
B --> M[Alexandria MCP]
D --> M
M --> U[Codex, ChatGPT, or another MCP client]Check out the Knowledge repository at the exact commit you want to serve.
Build the LanceDB index over its committed
.mdand.txtfiles.Prepare a data block containing the source snapshot, index, embeddings, model files, and release receipt.
Build the code-only Alexandria image and run it with the data block mounted at
/dataread-only.Connect a local client through stdio, or expose Streamable HTTP behind HTTPS and OAuth for remote clients.
The running service does not query GitHub. It searches the attached data block, so GitHub credentials never enter the image or the serving process.
The detailed commands are in Getting started. The complete release and replacement procedure is in Release workflow.
Related MCP server: data-olympus MCP server
Quick local run
After creating a verified data block at /path/to/alexandria-data:
docker build -t alexandria:2026.09.20 -f docker/Dockerfile .
docker run --rm -i \
--network none \
--read-only \
--mount type=bind,src=/path/to/alexandria-data,dst=/data,readonly \
alexandria:2026.09.20 serve --data /dataThe default read-only MCP surface is:
Tool | Purpose |
| Semantic retrieval |
| Literal ripgrep search |
| Read a bounded document range |
| Report the served generation and block state |
| List the authorized document tree and hashes |
submit_inbound is an explicit optional feature. It requires a writable
/inbound mount and --enable-inbound; the committed layer remains read-only.
Updating a deployment
There is no automatic GitHub-to-production refresh yet. Each update is a reviewable release: pull the new Knowledge commit, rebuild and verify the data block, build an image when runtime code changed, smoke-test the pair, and replace the service while retaining the previous pair for rollback.
Where to read next
Internal versions use CalVer YYYY.MM.DD. The current verified Python baseline
is 3.12, and the product uses the pi.alexandria PEP 420 namespace.
This server cannot be deployed
Maintenance
Related MCP Connectors
Read-only MCP server for the OrchestKit docs: full-text search + Markdown fetch. No auth.
Serve a folder of Markdown notes as an MCP server: hybrid search, reading, and sourced answers.
OAuth-protected, read-only-by-default MCP server for provenance-labeled QuillCaddie project memory.
Agent-native MCP server over the public saagarpatel.dev corpus. Read-only, stateless.
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceServes markdown knowledge from a git repository over MCP, providing tools to list, search, and retrieve documents.-
- AlicenseAqualityAmaintenanceProvides a single-writer MCP server for a governance-grade knowledge base of markdown documents with version control and query capabilities.49662 PyPI26Apache 2.0
- AlicenseNot gradedqualityDmaintenanceTurns a git-backed markdown directory into an MCP-compatible knowledge server, enabling any MCP client to read, search, ingest, and maintain a persistent wiki that compounds across sessions.MIT
- AlicenseNot gradedqualityAmaintenanceServes local Markdown wikis as a read-only knowledge source with MCP tools for agent context retrieval.130 PyPI6Apache 2.0