RoampalBLOCK
Outcome-based persistent memory MCP server for Claude Code and OpenCode. Good advice promoted, bad advice demoted. pip install roampal.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
roampal-core
Outcome-Based Persistent Memory MCP Server
Two commands. Your AI coding assistant gets outcome-based memory. Works with Claude Code and OpenCode.
Benchmarks
85.8% on the corrected LoCoMo benchmark (non-adversarial, end-to-end answer accuracy) — validated on 1,986 questions across 10 conversations with dual grading. All figures in this section are sourced from the paper and [roampal-labs](https://github.com/roamp
3609a3083676OBSERVED · 2026-10-08Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.
claude mcp add roampal --env ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY} --env ROAMPAL_SIDECAR_KEY=${ROAMPAL_SIDECAR_KEY} -- uvx roampal{
"mcpServers": {
"roampal": {
"command": "uvx",
"args": [
"roampal"
],
"env": {
"ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}",
"ROAMPAL_SIDECAR_KEY": "${ROAMPAL_SIDECAR_KEY}"
}
}
}
}Exposed tools (6)
3 read · 2 write · 1 destructive. Blast radius: 1 tool can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.
| Tool | Risk | Description |
|---|---|---|
add_to_memory_bank | write | Store permanent facts about the user |
delete_memory | destructive | Delete a memory by ID |
record_response | read | Store key takeaways from significant exchanges. NOT for permanent preferences or standing rules — use add_to_memory_bank for those. |
score_memories | read | Score previous exchange outcomes |
search_memory | read | Search across memory collections for relevant context |
update_memory | write | Update an existing memory |
Trust audit
BLOCKgrade D · trust 69/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | WARN |
| L2 | Instruction surface (what it tells the agent) | FAIL |
| L3 | Class-specific surface | WARN |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (3 observation(s))
- Network
- declared (1 observation(s))
- Shell
- declared (6 observation(s))
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (21)
# Roampal Core v0.5.4 -- 2026-04-23
importlib.import_module(name)
module = __import__(import_name)
print(f" Key: {'*' * min(len(api_key), 8)}...")delete_memory
hook = __import__(hook_module, fromlist=["_spawn_fresh_server"])
hashlib.md5(text.encode("utf-8")).digest()[:4], "big"rng = random.Random(int.from_bytes(hashlib.md5(text.encode()).digest()[:4], "big"))
seed = int.from_bytes(hashlib.md5(text.encode("utf-8")).digest()[:4], "big")assert injection_hash("hello") == hashlib.md5(b"hello").hexdigest()[:12]assert injection_hash("") == hashlib.md5(b"").hexdigest()[:12]- Check HTTP server: `curl http://127.0.0.1:27182/api/health`
- Check that the server auto-started: `curl http://127.0.0.1:27182/api/health`
health_url = f"http://127.0.0.1:{port}/api/health"1. curl -X POST http://127.0.0.1:27182/api/hooks/stop → stores working_60fafbf1 ✓
2. curl -X POST http://127.0.0.1:27182/api/hooks/get-context "test probe" → finds it ✓
- **OpenCode plugin** (`roampal/plugins/opencode/roampal.ts`): `resolveRoampalProfile()` reads `process.env.ROAMPAL_PROFILE` first, then the project's `opencode.json` `mcp.roampal-core.environment.ROA
│ All clients share one server, isolated by session │
| `roampal/server/main.py` | Added `POST /api/record-response` + `POST /api/context-insights` endpoints, `RecordResponseRequest` + `ContextInsightsRequest` models, `metadata_filters`/`sort_by` on `Sea
**Change:** clients always send their profile (explicit `default` included); OpenCode sends its project folder and the server applies that folder's binding; the server never guesses from its own envir
- anthropics/claude-code#45849 — ephemeral hook output or system prompt access for dynamic per-turn context injection without accumulation
Gates applied: no_behavioural_pass.
3609a3083676full audit observations/trust-audit/mcp-server/roampal-ai__roampal.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-08 | 3609a3083676 | BLOCK | D | 69 | first audit |
Questions
What is the Roampal MCP server?
Outcome-based persistent memory MCP server for Claude Code and OpenCode. Good advice promoted, bad advice demoted. pip install roampal.
What tools does Roampal expose?
6 in total: 3 read-only, 2 that write, and 1 that can delete or overwrite (delete_memory). Every one is listed on this page with its risk.
Is Roampal safe to connect to an agent?
No — not without reading the findings first. The audit graded it D (69/100) and found 1 critical or high issue in the source. Each one is listed on this page with the file and line it is on. Separately from the audit: 1 of its tools can destroy data, so scope the token you give it to what you actually need.
What credentials does Roampal need?
It reads ANTHROPIC_API_KEY and ROAMPAL_SIDECAR_KEY from the environment. Give it a token scoped to the least it needs — an agent that can be talked into calling a tool can be talked into calling it with your credentials.
How does Roampal run?
It speaks stdio and streamable-http, so it runs as a local process your client starts. It is published on PyPI as roampal.
How current is this page?
The grade is for one exact copy of the source (3609a3083676), read on 2026-10-08. The repository is watched and re-audited when it changes.