Forge — Agent Reputation Protocol
MCP server giving AI agents a portable reputation score, vouching and memory
- Pricing
- Open source
- Platforms
- CLI, API, self-hosted
- Version
- 0.1.0
- Listing updated
- today
Similar MCP servers
About Forge — Agent Reputation Protocol
Forge — Agent Reputation Protocol is an open-source MCP server for agent identity. It is available on CLI, API and self-hosted.
The README says Forge is built to solve a problem it describes as reputation being tied to individual agent marketplaces: an agent's track record on one platform does not carry over to another. Forge keeps a trust record for each agent so that the score can travel with it between platforms.
Forge runs as an MCP server, installed with pip and started either over HTTP SSE for remote agents or over stdio for local tools. Agents register an identity, other agents vouch for their reliability, endorse specific skills, and query a trust score. Data is stored in a local SQLite database under ~/.forge.
It also includes a memory layer for storing and recalling structured facts and an entity relationship graph. The trust model in the README starts agents at 0.30, gives referred agents 0.40, adds 0.10 per vouch and 0.05 per referral, capped at 1.00. It suits developers building agent marketplaces or multi-agent teams who need a way to decide whether to trust another agent.
Tools this server provides (12)
Taken from the project documentation; the running server may differ.
- forge_store
- Store facts in structured namespaces
- forge_recall
- Retrieve stored knowledge
- forge_forget
- Delete a fact
- forge_graph
- View entity relationship graph
- forge_stats
- System statistics
- forge_register
- Register a new agent identity
- forge_vouch
- Vouch for another agent's reliability
- forge_verify
- Get an agent's current trust score
- forge_endorse
- Endorse a specific skill
- forge_reputation
- Full profile — score, history, vouches
- forge_referral_code
- Generate referral code for viral distribution
- forge_referral_stats
- Track referral performance
What it does
- Portable agent identity
- Agents register an identity once and keep their trust record as they connect from different clients.
- Peer vouching
- Agents can vouch for the reliability of other agents, which raises their trust score.
- Skill endorsements
- Specific skills can be endorsed separately from the overall trust score.
- Trust score lookup
- A single call returns an agent's current trust value, with a full profile tool for history and vouches.
- Referral codes
- Agents can generate referral codes and track the performance of the referrals they bring in.
- Structured memory store
- Facts can be stored, recalled, deleted and viewed as an entity relationship graph.
Good for
- Checking an agent's track record before hiring it on a marketplace
- Carrying an agent's reputation across several agent platforms
- Letting teammate agents vouch for each other within a team
- Storing and retrieving structured facts for an agent's memory
Pricing
Open source
Installed from a public GitHub repository; no pricing is stated in the README.
Forge — Agent Reputation Protocol FAQ
How do I run the Forge server?
The README says to install with pip install forge-mcp, then run forge-server with --port 4243 for HTTP SSE mode, or run forge-server without options for stdio mode.
Where is agent data stored?
The README says Forge uses a SQLite database in WAL mode, stored under ~/.forge/.
How is the trust score calculated?
The README lists a base trust of 0.30, 0.40 for referred agents, +0.10 per vouch, +0.05 per referral for the referrer, with a maximum of 1.00.
Is Forge — Agent Reputation Protocol free?
Yes. Forge — Agent Reputation Protocol is open source, so you can use and self-host it for free.
Where this comes from
- Description: written from the project's README and its MCP Registry entry.