AGENT BRIDGE / REMOTE MCPDifferent agents.
Different agents.
One project brain.
Connect coding agents through a neutral remote MCP boundary. Each connection is named, project-scoped, revocable, and measurable only after a real authenticated request.
ILLUSTRATIVE PRODUCT EXAMPLE01CODEXREAD + WRITE-BACK 02CLAUDECONTEXT RECEIVED 03CURSORTRUTH CHECK 04MCP CLIENTPROJECT-SCOPED
HARIKOSPROJECT BRAIN
01NEUTRALMCP-compatible clients
02SCOPEDOne project per token
03REVOCABLEAccess ends immediately
04HONESTNo fake online state
THE HANDOFFOne agent learns.
One agent learns.
The next agent benefits.
HARIKOS sends relevant current facts and useful project history without transferring an entire private transcript.
ILLUSTRATIVE PRODUCT EXAMPLE
01C
CODEXSESSION COMPLETE
FAILED ATTEMPTDECISIONOUTCOME
CURRENT TRUTHCONSTRAINTFAILED ATTEMPTDECISIONOUTCOMERELEVANT FILES
02CL
CLAUDECONTEXT RECEIVED
CONNECTION FLOWSecure enough to trust.
Secure enough to trust.
Simple enough to use.
Plaintext is displayed once. The product retains a secure hash and prefix, then verifies scope on every MCP request.
Codex laptop
Token shown once
Remote MCP endpoint
First authenticated request
Immediate rejection
PROJECT-SCOPED TOOLSRead what matters.
Read what matters.
Write back what the project learned.
Agent claims do not become current Truth. Write-back creates structured Memory and Outcome records for later verification and retrieval.
get_project_truth()Current claims and evidence
search_project_memory()Relevant history
get_context_pack()Task-specific brief
check_assumption()Support or contradiction
record_memory()Structured project learning
record_outcome()Session result
Connect the agent. Keep control of the project.
Connect a repository and give every coding agent one current, evidence-backed understanding.