Skip to content
Agentmart
Submit a tool

PrivacyFence

Needs the PrivacyFence app from privacyfence.eu: human approval and audit for AI access to your data

Connector by privacyfence · GitHub account verified by the MCP Registry

IndexedApache-2.0Local (stdio)★ 1
Write a review

Install

Works with Claude Desktop. Read the repository before you give any tool access to your data.

Download the bundle

MCP bundle (.mcpb): open it with Claude Desktop.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Needs the PrivacyFence app from privacyfence.eu: human approval and audit for AI access to your data

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this server can do and what it asks for, from its tool list, install config and code. Checked Oct 26, 2026.

Tools
4 read · 1 write · 1 delete · 2 not yet classified
Secrets
None asked for
Install scripts
None found
Runs as
Downloaded binary

Tools by what they do 8

Read from the source code; the running server may list different tools.

Read 4 · Look things up without changing anything.

  • privacyfence_begin_unattended_session
  • privacyfence_check_policy
  • privacyfence_list_policy
  • privacyfence_status

Write 1 · Create, change, send or run things.

  • privacyfence_create_upload_slot

Delete 1 · Delete, drop, revoke or wipe things.

  • privacyfence_end_unattended_session

Not yet classified 2 · We could not tell from the name and description.

  • privacyfence_await_approval
  • privacyfence_propose_policy_change

Network hosts it talks to 22

  • api.anthropic.com · found in code
  • api.atlassian.com · found in code
  • api.github.com · found in code
  • auth.atlassian.com · found in code
  • chatgpt.com · found in code
  • claude.ai · found in code
  • developers.google.com · found in code
  • docs.astral.sh · found in code
  • docs.getgrist.com · found in code
  • downloads.privacyfence.eu · found in code
  • github.com · found in install command, code
  • glama.ai · found in code
  • googleapis.com · found in code
  • googletagmanager.com · found in code
  • json.schemastore.org · found in code
  • login.salesforce.com · found in code
  • privacyfence.eu · found in code
  • sitemaps.org · found in code
  • slack.com · found in code
  • test.salesforce.com · found in code
  • vertexaisearch.cloud.google.com · found in code
  • workspace.slack.com · found in code

Files it touches

  • /Library/PrivacyFence/image
  • /opt/privacyfence
  • /usr/sbin/privacyfence-privilege-separation
  • /usr/share/privacyfence/installer/linux
  • /var/lib/privacyfence
  • /var/lib/privacyfence/authority/logs/audit
  • %LOCALAPPDATA%
  • %LOCALAPPDATA%\\PrivacyFence
  • %LOCALAPPDATA%\\Programs
  • %USERPROFILE%
  • Your home directory (~)
  • ~/.privacyfence
  • ~/.privacyfence/authority/logs/audit
  • ~/Downloads
  • C:\\Program
  • C:\\ProgramData\\PrivacyFence\\authority\\logs\\audit
  • C:\\Windows
  • Any path the agent passes to its tools

Tools (8)

Read from the source code; the running server may differ. Descriptions are the server’s own words; hidden characters are shown as [U+…].

  • privacyfence_await_approval

    Long-poll pending approvals from gated calls' {status: 'approval_pending', approval_id, ...} results; status only, never content. Before the first call for an approval, relay that result's message and url (binder_url if several) to the user -- never wait silently. If pending_count > 1, first issue your other ready gated calls, then pass all approval_ids in one call (one human pass). Keep timeout_seconds under your client's tool-call timeout. Returns {approval_id: status}: 'pending' (schedule a follow-up if you can, else call again), 'approved' (re-issue the ORIGINAL call with identical argumen…

  • privacyfence_begin_unattended_session

    Tell PrivacyFence this conversation is an unattended/scheduled Cowork run (e.g. a Routine firing on a schedule) with no human necessarily watching, for the rest of this connection. From then on, any gated tool call that isn't already covered by a configured auto-accept rule is denied immediately with a clear error, instead of PrivacyFence opening a native approval dialog that nobody will answer. Call this once at the start of a scheduled run, and pair it with privacyfence_check_policy to plan which steps are safe to attempt. Never changes what auto-accepts, only what happens when nothing does.…

  • privacyfence_check_policy

    Before calling a gated tool, ask whether that exact call would auto-accept or need a human. Pass the connector, tool and args you're about to call. Returns {gate, verdict, matched_rule, matched_rule_id, reason, pii_gate_may_apply, pii_gate, pii_gate_detail, suggested_rules}. verdict: 'auto_accept' (the real call passes through identically), 'requires_review' (no rule can match these args) or 'unknown' (depends on fetched content). matched_rule_id (auto_accept only) is privacyfence_propose_policy_change's rule_id. suggested_rules: rules that would cover an uncovered call, {group, label, value, …

  • privacyfence_create_upload_slot

    Get a one-time URL to upload a local file to PrivacyFence, for a client without the PrivacyFence extension (e.g. Claude Code) -- use it when a tool's local_path/attachments parameter says PrivacyFence can't read your files directly. Returns {upload_id, upload_url, method: 'PUT', max_bytes, expires_at, example}: PUT the raw bytes to upload_url (e.g. the curl -T example); no Authorization header -- the URL is the credential. Then pass upload_id to the tool that needs the file (its upload_id parameter, e.g. drive_upload_file, or an 'upload:<upload_id>' attachments entry, e.g. gmail_*_with_attachm…

  • privacyfence_end_unattended_session

    Clear the unattended-session flag set by privacyfence_begin_unattended_session for this connection, restoring normal interactive approval behavior. Call this when a scheduled run finishes. Not strictly required -- the flag also clears automatically when the connection closes -- but call it if this connection might be reused afterward for something interactive. reason: one sentence on why the unattended session is ending now -- logged the same way as privacyfence_begin_unattended_session's.

  • privacyfence_list_policy

    List every configured auto-accept rule, plus the scope catalogue privacyfence_propose_policy_change accepts. Returns {rules, scope_groups}. Each rule has: id (pass as privacyfence_propose_policy_change's rule_id to update or remove it), sentence (human-readable), connector, scope_type, value, operations (internal keys, informational), verbs ({verb, family} pairs; family is 'read', 'write', 'send' or 'destructive'), conditions, and covered_tools (every tool name the rule can auto-accept). Each scope_groups entry has an id (pass as group), the verbs that scope can govern (pass a subset as verbs;…

  • privacyfence_propose_policy_change

    Propose adding, updating or removing an auto-accept rule (a scope plus allowed verbs). ALWAYS blocks on a human-approved dialog. If declined, or in an unattended session, it errors -- never assume success. Call privacyfence_list_policy first: group, verbs and rule_id must be listed, not guessed; a verb not in that group's scope_groups entry is rejected before any popup. operation='add'/'update' need group (a scope_groups id, e.g. 'drive.folder'), value (the resource's id -- folder id, channel ID, project key -- not its name or title, except label names and file types; a non-id is refused; omi…

  • privacyfence_status

    Check whether THIS PrivacyFence install is set up: call it before the first PrivacyFence-governed action in a conversation, or when asked why a connector (gmail_*, ...) is missing. An empty or partial tool list means connectors aren't authenticated yet, NOT that PrivacyFence is irrelevant -- this is the one tool guaranteed to exist even when every other tool is missing. Returns {mode ('local'/'org'), setup_complete (any connector authenticated), connectors [{name, enabled, authenticated, blocked_by: null, 'no_org_config', 'not_authenticated' or a reason}], next_step, message, sign_in_url (alwa…

Signals

Stars
1
Forks
0
Open issues
4
Last commit
Oct 23, 2026 (5 days ago)
Last release
Oct 25, 2026 (3 days ago)

From GitHub (fetched Oct 24, 2026).

Details

Registry name
io.github.privacyfence/privacyfence
Version
6.8.2
Auth
No auth
License
Apache-2.0
First seen
Oct 24, 2026
Last checked
Oct 26, 2026
Cite this page

PrivacyFence MCP server. Agent Mart. https://agentmart.dev/connectors/privacyfence-privacyfence (last updated 2026-10-26)