Public roadmap for MCPAtlas. Aligned with MCP_BLUEPRINT.md and DEEP_PLAN.md.
Goal: Installable, usable MCP server with core data and search.
- Core MCP server (STDIO, SSE, Streamable HTTP)
- 15 tools: search, get_project, compare, alternatives, health, suggest_stack, trends, graph, get_good_first_issues, get_migration_path, get_issue_context
- Resources and prompts
- Docker image and release workflow (multi-arch binaries + container + SBOM)
- Documentation (in-app docs at /docs), CONTRIBUTING, SECURITY, GOVERNANCE
- Homebrew formula (template; tap TBD)
- CI: site (landing + in-app docs) deploy via Pages workflow (enable Pages from Actions in repo settings)
- Tech debt:
min_starsandlanguagefilters wired in search handler + exposed in tool schema - Tech debt: protocol version negotiation (returns client's requested version)
- Tech debt:
logging/setLevelwired to tracing-subscriber reload handle
Goal: Semantic search, health scoring, richer data, CNCF Sandbox application.
- Storage trait abstraction (GraphBackend, CacheBackend, VectorBackend in
mcp-atlas-data;ProjectGraphimplementsGraphBackend) - SurrealDB integration for graph (optional
--graph-backend surreal, featuregraph-surrealdb; embedded kv-mem) - Qdrant + embeddings, hybrid BM25 + vector search (
mcp-atlas-vectorcrate;--qdrant-url+--embedding-api-base; RRF merge; optional featurevector-qdrant) - Artifact Hub integration (Helm packages per project;
--artifact-hub; pipeline step with rate limiting) - LLM summaries (Ollama/OpenAI-compatible), pipeline integration, content-hash cache skip, fallback
- Apply for CNCF Sandbox
Goal: Plugin ecosystem, new tools, operator, contributor growth.
- WASM plugin system (Extism):
PluginHosttrait,PluginRuntimestub, manifest; full host-function wiring TODO - Dynamic plugin tools:
DynamicTool,register_plugin_tool, tools/list merge, tools/call dispatch - Tools: get_good_first_issues, get_migration_path
- Contributor automation: PULL_REQUEST_TEMPLATE.md, CODEOWNERS, plugin-development docs
- Built-in plugins (e.g. Helm analyzer)
- Kubernetes operator (CRD, reconciliation)
Goal: Production-grade, horizontal scaling, security hardening.
- Redis for shared state (sessions, cache), multi-pod scaling
- Helm chart (deployment, service, ingress, HPA, PDB; Redis option TBD)
- Helm hardening: values.schema.json, Pod/container securityContext (runAsNonRoot, drop all caps, seccomp)
- Audit logging: structured tool_call events (tool, params_hash, status, latency_ms) to stderr; AuditLogger trait + StderrAuditLogger
- RBAC, optional auth
- SLSA Level 3 provenance, cosign signing
- Performance tuning (persistent Tantivy index, search cache)
Goal: Token-efficient issue resolution for CNCF-related repos.
-
get_issue_contexttool: repo + issue number → compact resolution brief (title, summary, labels, suggested files, CNCF project match) - GitHub API integration for issues (reqwest, runtime
GITHUB_TOKEN) - CNCF project metadata attached when repo is in landscape
- File path extraction from issue body, issue type classification from labels
- Workflow refinement: paste issue URL → AI gets brief → resolves in <5 turns, <20K tokens
- Issue comments and linked PRs in context brief
| Metric | 3 mo | 6 mo | 12 mo |
|---|---|---|---|
| GitHub stars | 500 | 2,000 | 10,000 |
| Monthly downloads | 1K | 10K | 100K |
| Contributors | 5 | 20 | 50+ |
| MCP tools | 15 | 15+ | 20+ |
| Plugins | 0 | 3 | 15+ |
| CNCF status | — | Sandbox | Incubating prep |
For detailed week-by-week execution see MCP_BLUEPRINT.md §11.