You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 4374f3c
Browse filesBrowse the repository at this point in the historyBrowse files
fix(cli,mcp): name the MCP server objectstack in os dev's banner, from the constant the served SKILL.md reads (#25425)
<!-- claude-projects-attribution: {"github_login":"os-sales"} -->
_Requested by **Sales** · [project
thread](https://claude.ai/code/project/chan_01D5bWFY3rFzgJcytD5gD4Uu?thread=cmsg_01D5bWFY3rFzgJcytD5gD4UuDWJsBGQpadCH8JR2xZQ5re)_
Part of #25341 (the `domain:cli` half: the boot banner and the served
SKILL.md). The console's Connect an Agent snippets are
objectstack-ai/objectui#13357; this PR leaves that half alone.
Clause-②: no
Metadata-protocol: none
## What changed
The maintainer's ruling (hotcrm#2194 T2-U22, ruling record `6354968262`)
says to use one server name in both places. This PR uses the platform
constant **`objectstack`**:
- `packages/mcp/src/skill-md.ts` exports `OBJECTSTACK_MCP_SERVER_NAME =
'objectstack'`, re-exported from `@objectstack/mcp`. The served
SKILL.md's `claude mcp add --transport http NAME URL` now reads it.
- `packages/cli/src/commands/dev.ts`: `printMcpConnectHint` reads the
same constant and drops its `name` option. The call site no longer
passes `path.basename(process.cwd()) || 'objectstack'`, which is where
`hotcrm` came from.
## Why this name and not the app's declared name
The issue allowed either the app's name or the deployment's. Today no
producer reads a declared app name:
- The banner used the **folder** name.
- The skill route (`packages/runtime/src/domains/mcp.ts`) calls
`renderSkill({ mcpUrl })` and has no app identity in reach. The skill is
generic across every app a key can reach (ADR-0036 Amendment C).
- objectui#13357 records that discovery serves no server name for the
console to read.
Using the app's name would need a new served field in both the skill
route and discovery. That is a contract addition, out of scope for an S
card. The console snippets already say `objectstack`, so with this PR
all three surfaces agree, and objectui#13357 has nothing to change.
## Pins
- `packages/cli/src/commands/dev-mcp-connect-hint-origin.test.ts`, new
block "the Connect row names the server the way the served SKILL.md
does":
- drives both producers (`printMcpConnectHint` and
`renderSkillMarkdown`) and asserts that each `claude mcp add` names
`OBJECTSTACK_MCP_SERVER_NAME`;
- a source pin asserts that dev.ts imports the constant and has no
`basename(process.cwd())`.
- The existing Connect-row expectations now read `objectstack`.
- `packages/mcp/src/skill-md.test.ts`: the skill's connect command names
the exported constant.
## Verification (head `ed39dfa362`)
- Build: `pnpm --filter '@objectstack/cli^...' build`. VERDICT
command-exit 0.
- `@objectstack/mcp`: `pnpm --filter @objectstack/mcp test` passed (42
files, 533 tests). Its typecheck is green.
- `@objectstack/cli`: `pnpm --filter @objectstack/cli build`, then
`vitest run --project unit` passed (326 files, 4804 tests). Its
typecheck is green.
- The integration layer is left to CI. No integration-tier or
spawn-entry file changed.
- `pnpm lint`: VERDICT command-exit 0.
- `dispatch-gates --ran`: 67 derived, 65 run with exit 0, 2 NOT
MEASURED, 0 UNRUN. The two NOT MEASURED gates:
- `check:dual-build-cjs-loads`
- `check:i18n-coverage`
Both exited 3 with PREREQUISITE NOT MET: they need packages outside this
closure built, and this local run did not build them. CI runs both on a
fresh build.
## Acceptance notes
- Docs examples (`content/docs/ai/connect-mcp.mdx`,
`content/docs/getting-started/build-with-claude-code.mdx`, the blog
post) name the server `my-app` / `support-desk`. In a user-typed command
that name is the user's own choice, so this PR leaves them as they are.
- `MCPServerRuntime`'s default handshake name (`mcp-server-runtime.ts`,
`name: 'objectstack'`) is a separate literal with the same value,
outside this card's file surface.
- If you registered the server under the folder name from an earlier
banner, that entry keeps working. The changeset gives the one-line
cleanup.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_01CxBKycaiUpA1yGx1XyoGqX
---
_Generated by [Claude
Code](https://claude.ai/code/session_01CxBKycaiUpA1yGx1XyoGqX)_
Co-authored-by: Claude <noreply@anthropic.com>
fix(cli,mcp): `os dev`'s boot banner and the served SKILL.md name the MCP server the same way
7
+
8
+
Clause-②: no
9
+
Metadata-protocol: none
10
+
11
+
`os dev`'s "connect a coding agent" block printed `claude mcp add --transport http FOLDER URL`, where FOLDER was the basename of the directory `os dev` ran in. The served `SKILL.md` (`GET /api/v1/mcp/skill`) and the console's Connect an Agent snippets name the server `objectstack`. So one endpoint ended up in an agent's config under two names, depending on which instructions were followed.
12
+
13
+
Both producers now read one exported constant, `OBJECTSTACK_MCP_SERVER_NAME` (`'objectstack'`), from `@objectstack/mcp`. The banner's `Connect` row prints `claude mcp add --transport http objectstack URL`. If you registered the server under the folder name from an earlier banner, that entry keeps working. Remove it with `claude mcp remove FOLDER` if you also added it as `objectstack`.
14
+
15
+
`printMcpConnectHint` no longer takes a `name` option.
0 commit comments