Talaiz and the open-source MCP gateways
This is the honest version of a comparison page. Several open-source gateways are excellent, some are better funded than we are, and for pure connectivity most of them will get you there faster. Here is what each one does well, and where we think the remaining gap is.
The honest short answer. If you need a smart, fast, well-adopted proxy between your agents and your tools, and your security requirement ends at “authorize the call”, use one of the projects below. Choose Talaiz when the requirement is evidence: an audit chain you can hand to someone else, human approval enforced by the server, and a connector that keeps customer secrets inside the customer’s network.
The open-source field
Verified from public sources on 2026-09-12. Star counts and funding move quickly; treat them as a snapshot, not a ranking.
| Project | What it does well | What it does not cover |
|---|---|---|
| Obot MIT · ~1k stars |
Per-tool policies by user and group, IdP OAuth, an audit log. Well funded (~$35M seed) and moving fast. | No documented human-approval primitive, no cryptographic audit chain, no confirmed SCIM. |
| agentgateway Apache-2.0 · ~4.8k stars |
CEL authorization on tool name and target, guardrails, OIDC/JWT. Now under the Linux Foundation with a large community. | A transport-smart gateway: no approval workflow, no tamper-evident chain, no outbound-only connector holding local secrets. |
| docker/mcp-gateway MIT · ~1.5k stars |
Allowlists, OAuth, secret handling and tracing, with the distribution reach of Docker. | The full governance edition is the paid MCP Enterprise Gateway, so the open edition stops short of approvals and audit evidence. |
| Lasso mcp-gateway MIT · ~386 stars |
Tool policies plus PII redaction via Presidio. | The surrounding platform is closed source; no approval workflow and no cryptographic audit chain. |
| Boomi MCPX MIT · ~493 stars |
Tool-level access control with a real product team behind it after the Lunar.dev acquisition. | No SSO in the open edition; the project’s direction now follows a commercial platform. |
Where Talaiz is genuinely different
Four things, and we would rather be judged on these than on a feature checklist:
- One sealed audit event per call, chained per organization. Every executed call appends exactly one event to a SHA-256 hash chain, and retention is anchored by a signed checkpoint held outside the database. Tampering, reordering or truncation is detectable, and an append failure fails the call closed — no unaudited result ever reaches an agent.
- Approval is a policy decision, not a UI feature. A tool marked require_approval blocks the call until a person approves or denies it, and that decision lands in the same chain as the call it authorized.
- The connector is outbound-only and fails closed. It dials out, never listens, keeps customer credentials on the customer’s machine, and stops executing tools the moment it cannot reach the control plane. It is one small binary with no heavy dependencies.
- Multi-tenant and operator-ready out of the box. Per-organization isolation, SSO via OIDC with per-org providers, SCIM provisioning, plan quotas, audit export to webhook or S3, and a published price.
What we do not claim
- We are not the fastest path to a working proxy. Several projects above are.
- We are not the biggest community. agentgateway and Docker are far ahead of us.
- We are not cheaper than free. The open source is free to self-host, and the paid part is the managed control plane.
- We are not a tunnel. If connectivity is your whole problem, a tunnel is a solved commodity.
The honest tie-breaker
Ask one question of any gateway, including ours: if someone with database access rewrote yesterday’s log, how would I know? If the answer is “I trust the database”, the product is a proxy with a log file. If the answer is a hash chain plus a signed anchor you can verify on demand, it is evidence.
We may be wrong about your project. Every claim above was read from public sources on 2026-09-12 and may already be out of date. If we have mischaracterized your work, tell us and we will correct it. Comparing ourselves favourably is not worth stating something false.