From c2e670c9b8fec695ad3925471e4a5956a84683ea Mon Sep 17 00:00:00 2001 From: Tim Crooker Date: Sun, 14 Jun 2026 13:03:19 -0400 Subject: [PATCH] chore: changeset for unified MCP layer (core+cli minor, templates patch) Co-Authored-By: Claude Opus 4.8 (1M context) --- .changeset/unified-mcp-layer.md | 14 ++++++++++++++ 1 file changed, 14 insertions(+) create mode 100644 .changeset/unified-mcp-layer.md diff --git a/.changeset/unified-mcp-layer.md b/.changeset/unified-mcp-layer.md new file mode 100644 index 0000000..66fba53 --- /dev/null +++ b/.changeset/unified-mcp-layer.md @@ -0,0 +1,14 @@ +--- +"@timothycrooker/ai-context-core": minor +"@timothycrooker/ai-context-cli": minor +"@timothycrooker/ai-context-templates": patch +--- + +Add a unified MCP layer — MCP is now a third generated primitive alongside context and skills. + +Declare MCP servers once in `.ai/mcp.json`; `ai-context build` fans them out to each agent client's native config (Claude `.mcp.json`, Codex `.codex/config.toml`). Servers can carry backing: a linked skill (auto by co-name or explicit) and a one-line catalog entry in `AGENTS.md`/`CLAUDE.md` so an agent gets the tool and the knowledge to use it. + +- `project`-scope servers are committed; `user`-scope servers install per-machine via `ai-context mcp install --user`. +- Secrets stay as `${VAR}` references (resolved from `.ai/secrets.local.env`); `ai-context verify` fails on a credential literal in a generated config. +- New CLI: `ai-context mcp list | install --user | setup`. +- v1 ships `claude` + `codex` adapters behind a pluggable registry.