Why agent security needs a tool boundary
An AI agent does more than generate text. It can select tools, pass arguments, read external content, and cause changes in connected systems. That makes the connection between the model and its tools a security boundary, not merely an integration detail.
Traditional application controls remain necessary, but they do not explain why a model selected a tool or whether the call matched an approved policy. Bindfort is being built as an enforcement and evidence layer between MCP-based agents and the servers they use.
- Review installed dependencies rather than only declared package names.
- Apply allow and deny policy before an MCP tool call reaches its upstream server.
- Record a tamper-evident receipt that connects the call to its policy decision.
- Keep telemetry and evidence under the operator’s control where possible.
What Bindfort can establish today
The current verified path is deliberately narrow: the installed-tree scanner is publicly downloadable, while local allow and deny decisions, receipt generation, and receipt verification are available through guided technical evaluation rather than a self-serve production gateway.
Runtime isolation, broad compatibility packaging, automated incident response, and production alert routing remain roadmap work. Bindfort labels those boundaries so security teams can distinguish verified behavior from product direction.
Product status
Verified today and clearly separated from roadmap
- Public, account-free installed dependency-tree scanning for local MCP server directories.
- Allow and deny policy decisions before controlled upstream execution.
- Tamper-evident receipt records and local receipt verification.
- Structured evidence for security, vendor-risk, and audit review.
- Packaged runtime containment profiles for untrusted MCP servers.
- Automated alert routing and production response workflows.
- Broader client, server, and transport compatibility coverage.
Frequently asked questions
Clear answers for evaluation
What is AI agent security?
AI agent security is the set of controls used to protect an agent’s identities, instructions, tools, data access, execution paths, and audit evidence.
Is Bindfort a general-purpose AI firewall?
No. Bindfort is focused on MCP tool calls, installed software risk, policy enforcement, and evidence. It does not claim to replace every model, data, or application security control.
Is Bindfort generally available?
The scanner-only CLI is publicly available without an account. The policy gateway and paid production packages remain guided while production packaging is completed.