tap-oss
Credential isolation for AI agents — stop leaking secrets to your agents. TAP is a credential proxy: agents never see secret values; optional human approval on sensitive calls. By human.tech (Apache-2.0).
Documentation
Tool Authorization Protocol (TAP)
Credential isolation, approval gating, and connector routing for AI agents.
This repository contains the code that is most useful for:
- auditing request routing and connector behavior
- debugging failed requests
- understanding approval flows
- improving connector-side request shaping
- contributing fixes to the core TAP experience
> [!WARNING]
> Self-hosting means you own the security of your credentials and signing keys.
> TAP keeps secrets out of your agents, but running it yourself puts the host
> hardening, key isolation, and correct policy-engine operation on you. It's a path
> for teams that are well versed in security. For everyone else the hosted version is
> strongly recommended: credentials sit in a hardware enclave we can't read into, with
> no ops to run. Start free at tap.human.tech.
Start Here
- `crates/tap-proxy/src/routing.rs` — how TAP resolves connector target shapes
- `crates/tap-proxy/src/placeholder.rs` — credential substitution and position validation
- `crates/tap-proxy/src/policy.rs` — approval policy enforcement
- `docs/` — full documentation including self-hosting guide
Included
- core proxy and storage crates
- Telegram and Matrix approval bots
- remote MCP server (`tap-mcp`)
- CLI
- docs (self-hosting, API reference, credential setup)
Not Included
- enclave deployment glue (CCE policy generation, release-policy automation, ARM templates, env config)
- production workflows and secret bootstrapping
- managed hosting operations glue
- the hosted dashboard UI source (a placeholder is shipped so the proxy compiles)
The enclave key-management source is included (`key_provider_enclave.rs`,
`kms_azure.rs`, `skr.rs`) — it's the custody model documented at
docs.tap.human.tech/security, and each
hosted release's enclave measurement is published in `measurements/`.
Hosted deployment and operational infrastructure are maintained separately from this repository.
Security
See `SECURITY.md` to report a vulnerability.
License
Apache-2.0: free to use, read, modify, and self-host. This repo is the
open-source TAP runtime (`tap-core`, `tap-proxy`, `tap-bot`, `tap-cli`, `tap-mcp`). The
hosted dashboard and managed-service deployment glue are proprietary and live in
a separate private repo.
Contributing
See `CONTRIBUTING.md`.
Testing
# Needs Postgres (default postgres://tap:tap@localhost:5434/tap, override with
# POSTGRES_DATABASE_URL). Isolated suites parallelize; env-mutating unit tests stay serial.
cargo test -p tap-core
cargo test -p tap-proxy --test integration --test e2e
cargo test -p tap-proxy -p tap-bot -p tap-cli --lib --bins -- --test-threads=1Frequently asked questions
What is tap-oss?
tap-oss is Credential isolation for AI agents — stop leaking secrets to your agents. TAP is a credential proxy: agents never see secret values; optional human approval on sensitive calls. By human.tech (Apache-2.0).
How do I install tap-oss?
Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.
Is tap-oss open source?
Yes — it is hosted on GitHub at https://github.com/holonym-foundation/tap-oss and has 12 stars.
Related MCP tools
Open-source AI agent firewall for MCP security and agent egress. Scans mediated HTTP, MCP, A2A, and WebSocket traffic for exfiltration, SSRF, and prompt injection, and emits mediator-signed action receipts: verifiable audit evidence from outside the agent.
an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM
Fast, local-first web content extraction for LLMs. Scrape, crawl, extract structured data — all from Rust. CLI, REST API, and MCP server.
Markdown knowledge graph — LSP for your editor, CLI + MCP memory for your AI agents
Search infrastructure for AI
YC (S26) | Open Computer History | Record your screen continuously locally and provide context to your agents (Claude, Codex, Openclaw, Hermes, Runner...)
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP