SAIHM
Sovereign, encrypted, persistent memory for AI agents. Eight MCP tools — remember, recall, forget (GDPR Art. 17 erasure), share, governance. Anchored on COTI V2 mainnet. Apache-2.0.
README
SAIHM MCP Server
Sovereign, encrypted, sharable, persistent memory protocol for AI agents.
v0.1.2 · Apache-2.0 · COTI V2 mainnet
What this is
A Model Context Protocol server that exposes eight tools any MCP-capable AI agent (Claude Code, Claude Desktop, custom agents) can call to gain a persistent, encrypted memory layer the user owns:
saihm_remember— store an encrypted memory cellsaihm_recall— retrieve and decrypt your memoriessaihm_forget— true cryptographic erasure (GDPR Art. 17)saihm_status— your protocol-runtime stats and storage tier dashboardsaihm_share/saihm_revoke_share— selectively share a memory with another agent or usersaihm_governance_propose/saihm_governance_vote— protocol governance via gSAIHM
Each tool forwards to a SAIHM operator endpoint that runs the full protocol stack on COTI V2 mainnet. The server itself holds no crypto, no storage, and no protocol runtime — those live behind the operator endpoint.
Install
npm install @saihm/mcp-server
# or run directly without install:
npx @saihm/mcp-server
Configure
The server needs two env vars:
SAIHM_ENDPOINT_URL=https://operator.example.com/saihm/v1
SAIHM_AUTH_HEADER=Bearer <token-issued-by-your-operator>
SAIHM_ENDPOINT_URL— the SAIHM operator endpoint. Operators publish their endpoint URLs at https://saihm.coti.global.SAIHM_AUTH_HEADER— theAuthorizationheader value the operator expects (typically aBearer <token>issued to you after key-bound enrolment). The server is authentication-agnostic and never transmits raw private keys; the operator's enrolment flow keeps your Wallet C key on your machine.
Place these in a .env file alongside the server (the .gitignore excludes
all .env* files from any future repo).
Wire into Claude Code
{
"mcpServers": {
"saihm": {
"command": "npx",
"args": ["@saihm/mcp-server"],
"env": {
"SAIHM_ENDPOINT_URL": "https://operator.example.com/saihm/v1",
"SAIHM_AUTH_HEADER": "Bearer <token>"
}
}
}
}
What gets persisted, where
The server itself persists nothing. The operator endpoint runs the full protocol stack: cells are encrypted under a per-cell DEK, sealed by a per-agent KEK, persisted to Filecoin, and audited on COTI V2 mainnet. See the operator's documentation for tier details.
Reporting engine
A reporting library is bundled as a sub-export, so operators can compose the eight MCP calls into bespoke reports with their own tooling (no extra dependency, no extra service):
import {
validateBespokeTemplate,
registerTemplate,
generateRegistryAttestation,
StubPublicRegistry,
InMemoryReportingRuntime,
GDPR_ART15_FIELDS,
REGISTRY_ATTESTATION_FIELDS,
type BespokeReportTemplate,
} from "@saihm/mcp-server/reporting";
What it covers
- Field universe (
FIELD_UNIVERSE) — 280 fields (262 framework + 18 ledger). Templates that project a field outside this set are rejected at validation. - Bespoke template schema — zod validator + universe-membership check + scope/cap enforcement.
- Authorization path validators — 4 paths:
public/self/operator-self/operator-for-downstream. - Receipt emission — 6 sub-kinds (
report_generated/report_rejected/template_registered/template_superseded/erasure_chain_broken/rate_limit_exceeded) under a stable HKDF receipt domain. - Framework smoke —
registry-attestation(public auth) for end-to-end plumbing verification.
Constraints
- Every
fieldProjections[]entry MUST be inFIELD_UNIVERSE. scope.customerIdHashes64-hex; max 10,000 per template.scope.timeRangewindow ≤ 366 days.fieldProjectionslength 1–200.framework∈ {gdpr-art-15,gdpr-art-17,soc2-t1,soc2-t2,iso27001,aml,audit-export,billing-history,registry-attestation}.format∈ {pdfa3,json,csv}.
Worked example
const template: BespokeReportTemplate = {
templateId: "acme-q1-summary",
templateVersion: 1,
operatorIdHash: "ab".repeat(32),
scope: {
customerIdHashes: ["cd".repeat(32)],
timeRange: { from: "2026-01-01T00:00:00Z", to: "2026-04-01T00:00:00Z" },
},
framework: "gdpr-art-15",
fieldProjections: [GDPR_ART15_FIELDS[0], GDPR_ART15_FIELDS[1]],
format: "pdfa3",
};
const v = validateBespokeTemplate(template);
if (!v.valid) throw new Error(v.errors.join(", "));
const runtime = new InMemoryReportingRuntime(); // replace with your audit-ledger runtime
const reg = await registerTemplate(template, runtime);
if (reg.ok) console.log("registered:", reg.templateHash);
In production, replace InMemoryReportingRuntime with a runtime that persists audit payloads to your operator's audit ledger. Operators who inject signature verifiers should use pure-crypto libraries (@noble/curves for EIP-712, @noble/post-quantum for FIPS 204 ML-DSA) — the package itself bundles no EVM tooling.
Security
The server enforces a small set of defaults so misconfiguration cannot leak the Authorization header in transit:
- HTTPS-only endpoints.
SAIHM_ENDPOINT_URLmust usehttps://. Plainhttp://is rejected at construction time, except for127.0.0.1andlocalhost(so a local operator endpoint works during development). - Per-call abort window. Each request runs under an
AbortControllerthat aborts after 30s, preventing a hung endpoint from starving the MCP server. - Response-size cap. Responses whose
Content-Lengthexceeds 16 MB are rejected before deserialisation. - No header echo.
Authorizationis never included in thrown error messages or stdout. - No filesystem reads. The package never reads from disk; configuration flows entirely through env vars.
- Zero EVM tooling. No
ethers, noeth_*, no Solidity. If operators inject signature verifiers viaAuthVerifiers, they should use pure-crypto libraries (@noble/curves,@noble/post-quantum).
Trust model: this client trusts whatever endpoint the operator configures. Cell IDs, audit anchors, and report receipts returned from that endpoint are surfaced to the agent verbatim — operators are the authority for content shown via saihm_recall. Verifying receipts against COTI V2 mainnet anchors is out of scope for this server; consume the cellId and auditCellId fields and verify against your own SAIHM mainnet read path.
For distribution integrity, the SAIHM publisher signs releases via npm sigstore provenance; verify with npm view @saihm/mcp-server --json | jq .dist and npm audit signatures.
License
Apache-2.0 — see LICENSE.
Project
- Site: https://saihm.coti.global
- Issue tracker: see homepage for current contact channel.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。