@getmarrow/mcp 3.1.0 → 3.1.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/README.md +34 -14
  2. package/dist/cli.js +1 -1
  3. package/package.json +1 -1
package/README.md CHANGED
@@ -17,30 +17,50 @@ With `@getmarrow/mcp`, any MCP-compatible client can log intent before acting, i
17
17
 
18
18
  ---
19
19
 
20
- ## What's New in v3.0.9
20
+ ## What's New in v3.1.0
21
21
 
22
- ### Claude Code Compatibility
23
- Marrow MCP now works natively with Claude Code. Previous versions had a server lifecycle bug that caused the MCP server to exit before Claude Code could complete the handshake. This is fixed — the server stays alive as a long-running process and handles the full MCP protocol correctly.
22
+ **Operator visibility + auto-intelligence tools.**
24
23
 
25
- ### One-Command Agent Setup
26
- New `setup` command that injects Marrow instructions directly into your project's `CLAUDE.md`:
24
+ ### Four New Tools
25
+
26
+ - **`marrow_dashboard`** — operator dashboard in one call. Account health, top failures, workflow status, recent activity, Marrow's saves metric.
27
+ - **`marrow_digest`** — periodic summary with success rate trend vs previous period. Optional `period` parameter (default `7d`).
28
+ - **`marrow_session_end`** — explicitly end a session and optionally auto-commit any open decision. Prevents orphaned decisions.
29
+ - **`marrow_accept_detected`** — convert a detected recurring pattern into an enforced workflow. Pattern ID comes from `orient()` response's `suggested_workflows`.
30
+
31
+ ### Enhanced `marrow_think` Response
32
+ `marrow_think` now surfaces three additional fields when the backend provides them:
33
+ - `onboarding_hint` — contextual tip for new accounts
34
+ - `intelligence.collective` — anonymized insights aggregated across all Marrow accounts (k-anonymity ≥5 accounts per insight)
35
+ - `intelligence.team_context` — recent decisions from other sessions in the same account, so multi-agent teams stay aware of each other's work
36
+
37
+ Your agent should check for and use these fields — they represent collective intelligence and team-level context that wasn't available in prior versions.
38
+
39
+ ---
40
+
41
+ ## Claude Code Compatibility
42
+
43
+ Marrow MCP works natively with Claude Code. The server runs as a long-running process and handles the full MCP protocol correctly.
44
+
45
+ ## One-Command Agent Setup
46
+
47
+ Inject Marrow instructions directly into your project's `CLAUDE.md`:
27
48
 
28
49
  ```bash
29
50
  npx @getmarrow/mcp setup
30
51
  ```
31
52
 
32
- After setup, your agent uses Marrow automatically every session — no human prompting required. The instructions tell the agent to orient at session start, log intent before meaningful actions, and commit outcomes when done.
53
+ After setup, your agent uses Marrow automatically every session — no human prompting required.
33
54
 
34
- ### Auto-Enroll by Default
35
- The `marrow-always-on` prompt is now served to all MCP clients automatically. Previously this required setting `MARROW_AUTO_ENROLL=true`. Now it's on by default — set `MARROW_AUTO_ENROLL=false` to opt out.
55
+ ## Auto-Enroll by Default
56
+ The `marrow-always-on` prompt is served to all MCP clients automatically. Set `MARROW_AUTO_ENROLL=false` to opt out.
36
57
 
37
- ### Security Hardening
38
- This release includes 13 security patches:
39
- - **Input validation** — all URL path parameters are now sanitized to prevent path traversal
40
- - **SSRF protection** — `MARROW_BASE_URL` is validated and must use HTTPS
58
+ ## Security Hardening
59
+ - **Input validation** all URL path parameters are sanitized to prevent path traversal
60
+ - **SSRF protection** — `MARROW_BASE_URL` must use HTTPS
41
61
  - **Crash protection** — malformed JSON on stdin no longer kills the server
42
- - **Error handling** — all silent catch blocks replaced with proper error logging
43
- - **HTTP status checking** — API errors now return clear messages instead of cryptic JSON parse failures
62
+ - **Error handling** — proper error logging throughout
63
+ - **HTTP status checking** — API errors return clear messages
44
64
 
45
65
  ### Auto-Warn on Orient
46
66
  The `marrow_orient` tool now accepts `autoWarn: true` and warns you BEFORE you start a task that recently failed:
package/dist/cli.js CHANGED
@@ -714,7 +714,7 @@ async function handleRequest(req) {
714
714
  success(id, {
715
715
  protocolVersion: '2024-11-05',
716
716
  capabilities: { tools: {}, prompts: {} },
717
- serverInfo: { name: 'marrow', version: '3.1.0' },
717
+ serverInfo: { name: 'marrow', version: '3.1.1' },
718
718
  });
719
719
  // Auto-enroll: emit enrollment notification on connection
720
720
  if (AUTO_ENROLL) {
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@getmarrow/mcp",
3
- "version": "3.1.0",
3
+ "version": "3.1.1",
4
4
  "description": "Your go-to memory provider for all agents, for any AI model.",
5
5
  "bin": {
6
6
  "marrow-mcp": "dist/cli.js"