Feat: add wiki-memory-tool MCP service for multi-agent wiki knowledge

Makefile

@@ -29,7 +29,7 @@ test-a2a:
 	done
 
 # Directories under mcp/ to skip in test-mcp
-TEST_MCP_SKIP := mcp/exgentic_benchmarks
+TEST_MCP_SKIP := mcp/exgentic_benchmarks mcp/wiki_memory_tool
 
 # Verify all of the MCP example Docker images can be built
 # (Optional KAGENTI_DOCKER_FLAGS for docker build, e.g. --no-cache or --load)

mcp/wiki_memory_tool/.gitignore

@@ -0,0 +1,10 @@
+# Python-generated files
+__pycache__/
+*.py[oc]
+build/
+dist/
+wheels/
+*.egg-info
+
+# Virtual environments
+.venv

mcp/wiki_memory_tool/.python-version

@@ -0,0 +1 @@
+3.14

mcp/wiki_memory_tool/DEMO.md

@@ -0,0 +1,154 @@
+# Wiki Memory Tool Demo
+
+A quick walkthrough of the wiki CLI — login, query, discover, and verify.
+
+## 1. Login
+
+```bash
+$ kwiki login
+==================================================
+  GitHub Device Authorization
+==================================================
+
+  1. Open this URL in your browser:
+
+     https://github.com/login/device
+
+  2. Enter this code:
+
+     ABCD-1234
+
+  Code expires in 15 minutes.
+==================================================
+
+Waiting for authorization.....
+
+Logged in as aslom
+Groups: kaslomorg/ml-writers, kaslomorg/platform-admins
+```
+
+## 2. Check Identity
+
+```bash
+$ kwiki whoami
+User:   aslom
+Status: valid (expires in 6d 23h)
+Groups: kaslomorg/ml-writers, kaslomorg/platform-admins
+Server: https://wiki-memory-service-team1.apps.ykt1.hcp.res.ibm.com
+
+Access:
+  ai: read, write, admin
+    read  <- *
+    write <- github:team:kaslomorg/ml-writers
+    admin <- github:user:aslom
+  security: read, write
+    read  <- github:org:kaslomorg
+    write <- github:team:kaslomorg/platform-admins
+```
+
+## 3. Query Existing Pages
+
+```bash
+$ kwiki query list-topics
+  ai (7 pages)
+  security (0 pages)
+  ml (0 pages)
+
+$ kwiki query list-pages ai
+  evaluation.md
+  fine-tuning.md
+  rag-patterns.md
+  transformers.md
+
+$ kwiki query search ai "attention mechanism"
+  ai/transformers.md (score=0.1652)
+    Self-attention mechanisms for parallel sequence processing.
+
+$ kwiki query read ai transformers.md
+---
+tags: [paper, architecture]
+---
+# Transformer Architecture
+
+Self-attention mechanisms for parallel sequence processing.
+...
+```
+
+## 4. Discover — Write a New Page
+
+```bash
+$ kwiki discover novelty ai "Mixture of Experts" "Sparse gating for efficient model scaling"
+NOVEL: No sufficiently similar content found
+
+$ kwiki discover write ai mixture-of-experts.md --content "$(cat <<'EOF'
+---
+tags: [paper, architecture, scaling]
+---
+# Mixture of Experts
+
+Source: [Switch Transformers](https://arxiv.org/abs/2101.03961) (Fedus et al., 2021)
+
+## Summary
+
+Mixture of Experts routes each token to a subset of specialist sub-networks,
+enabling model scaling without proportional compute increase.
+
+## Key Ideas
+
+- Sparse gating: each token activates only top-k experts
+- Load balancing loss prevents routing collapse
+- Capacity factor limits tokens per expert
+
+## Results
+
+- Switch Transformer: 7x speedup over T5-Base at same compute
+- Mixtral 8x7B: competitive with GPT-3.5 using 2 active experts per token
+EOF
+)"
+Written: ai/mixture-of-experts.md by discovery-agent
+Suggested links:
+  ai/transformers.md (score=0.0923)
+  ai/fine-tuning.md (score=0.0412)
+```
+
+## 5. Verify It Was Added
+
+```bash
+$ kwiki query search ai "mixture of experts sparse gating"
+  ai/mixture-of-experts.md (score=0.2841)
+    Mixture of Experts routes each token to a subset of specialist sub-networks.
+
+$ kwiki query read ai mixture-of-experts.md
+---
+tags: [paper, architecture, scaling]
+---
+# Mixture of Experts
+
+Source: [Switch Transformers](https://arxiv.org/abs/2101.03961) (Fedus et al., 2021)
+...
+
+$ kwiki query tags ai
+  paper: 3 pages
+  architecture: 2 pages
+  scaling: 1 pages
+  retrieval: 1 pages
+  technique: 1 pages
+
+$ kwiki query backlinks ai mixture-of-experts.md
+  (none yet)
+```
+
+## 6. Optional — Initialize GitHub Pages
+
+```bash
+$ kwiki admin init-pages
+GitHub Pages initialized (6 files):
+  _config.yml
+  index.md
+  _layouts/default.html
+  _layouts/page.html
+  _includes/nav.html
+  assets/css/style.css
+```
+
+After ~60 seconds, browse: https://kaslom.github.io/kagenti-wiki-research/

mcp/wiki_memory_tool/Dockerfile

@@ -0,0 +1,25 @@
+FROM quay.io/fedora/python-314@sha256:381b2b1135f4c506cc1287223eff3ec6ff585638297ecb141667203a89116375
+
+USER root
+RUN dnf install -y git && dnf clean all
+COPY --from=ghcr.io/astral-sh/uv:latest /uv /usr/local/bin/uv
+
+WORKDIR /app
+COPY pyproject.toml /app/
+COPY wiki_service.py /app/
+COPY run_local.py /app/
+COPY mcp_server.py /app/
+RUN uv sync --no-dev && chmod -R 755 /app/.venv
+
+RUN mkdir -p /tmp/uv-cache && chmod 755 /tmp/uv-cache
+USER 1001
+
+ENV UV_CACHE_DIR=/tmp/uv-cache
+ENV UV_LINK_MODE=copy
+ENV WIKI_ROOT=/data/wiki
+ENV ACL_FILE=/config/acl.yaml
+ENV SPIFFE_TRUST_DOMAIN=kagenti.example.com
+
+EXPOSE 8000
+
+CMD ["/app/.venv/bin/uvicorn", "wiki_service:app", "--host", "0.0.0.0", "--port", "8000"]