MCP server integration + per-conversation vault binding by RaghavSood · Pull Request #2 · vultisig/agent-backend

RaghavSood · 2026-02-20T18:22:07Z

Summary

Adds an MCP (Model Context Protocol) JSON-RPC 2.0 client that connects to the Vultisig MCP server, discovers tools (e.g. get_eth_balance, get_token_balance, set_vault_info), and exposes them to Claude alongside existing built-in tools
Introduces a set_vault built-in tool that stores vault keys (ecdsa_public_key, eddsa_public_key, chaincode_hex) per-conversation in Postgres, so the MCP server can derive addresses automatically
On every message request, automatically primes the MCP session with the active vault — no manual re-sending of keys by the user

Why this is needed

Users have multiple vaults and switch between them. When Claude calls MCP tools like get_eth_balance, the MCP server needs to know which vault's keys to use for address derivation. Without per-conversation tracking, the user would need to re-provide keys every time, or worse, the wrong vault's balance could be returned.

By binding vault keys to the conversation:

The user (or app) calls set_vault once when selecting a vault
Every subsequent message automatically primes the MCP session with those keys
If the user switches vaults mid-conversation, calling set_vault again updates both the DB and the MCP session
Claude sees the active vault in the system prompt, so it can reference the keys contextually

How it works

MCP client (`internal/mcp/client.go`)

Lightweight JSON-RPC 2.0 over Streamable HTTP, no external SDK
Initialize() → ListTools() at startup; tools cached in-memory with configurable TTL
Stale cache fallback: if a refresh fails, the last-known tools are still served
Session ID tracked via Mcp-Session-Id header
Structured logging at every interaction point for debuggability

Vault lifecycle

App sends vault keys → Claude calls set_vault → stored in agent_conversations
                                               → forwarded to MCP set_vault_info

Next message arrives → ProcessMessage loads conversation
                     → vault info found → auto-calls MCP set_vault_info
                     → Claude calls get_eth_balance → MCP derives address from vault keys

Configuration

Env var	Default	Description
`MCP_SERVER_URL`	(empty = disabled)	MCP server endpoint (e.g. `http://localhost:8888`)
`MCP_TOOL_CACHE_TTL_SECONDS`	`300`	How long to cache discovered MCP tools

Database

New migration adds three nullable columns to agent_conversations:

ecdsa_public_key TEXT
eddsa_public_key TEXT
chaincode_hex TEXT

Test plan

Set MCP_SERVER_URL=http://localhost:8888, start server — logs should show mcp tools loaded with tool count
Unset MCP_SERVER_URL, start server — no MCP logs, server works normally
Send a message with vault keys via set_vault tool — verify DB columns populated, MCP set_vault_info called
Send follow-up message — verify MCP session auto-primed (debug log: mcp session primed with vault info)
Call set_vault again with different keys — verify DB updated, new keys forwarded to MCP
Ask Claude "what tools do you have?" — should list both built-in and MCP tools
Trigger an MCP tool (e.g. ask for ETH balance) — verify MCP tool called and result returned

🤖 Generated with Claude Code

neavra

Left some comments! Please take a look

neavra · 2026-02-23T03:36:01Z

cmd/server/main.go

 	})

-	// Agent routes (authenticated)
-	agent := e.Group("/agent", server.AuthMiddleware)


Agent routes should be authenticated so that only vault owners can access conversations

neavra · 2026-02-23T03:50:40Z

internal/service/agent/tools.go

 }

+// SetVaultTool sets the active vault for this conversation.
+var SetVaultTool = anthropic.Tool{


Would it be better to do on creation of a new conversation? Since the conversation is tied to a vault we should be able to get the needed keys on conversation creation? Let me know if im misinterpreting something

Connects the agent backend to a Vultisig MCP server (optional, via MCP_SERVER_URL) so Claude can discover and call external tools like get_eth_balance and get_token_balance. Adds a set_vault built-in tool that binds vault keys (ECDSA, EdDSA, chaincode) to a conversation. On each request the MCP session is automatically primed with the active vault so address derivation happens server-side without the user repeating keys. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Add nix flake dev environment with Go, PostgreSQL, Redis, sqlc - Add Transaction type and extract transactions from MCP tool results - Remove redundant public key auth checks (already handled by middleware) - Rename chaincode_hex to chain_code for consistency Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Extract token search results from the find_token MCP tool and pass them as structured data in the API response so frontend apps can prompt users to add tokens/chains to their vault. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Two issues prevented the tokens field from appearing in API responses: 1. CallTool discarded text on IsError — when the MCP tool set IsError: true, CallTool returned a Go error and the text content was lost. Now returns a ToolError that carries the text, so executeTool can still pass it to trackToolResult for structured extraction. 2. trackToolResult assumed pure JSON — MCP tools may return multiple text content blocks (joined with \n) or mix descriptive text with JSON. The direct json.Unmarshal failed silently. Now uses extractTokens() which tries direct unmarshal first, then scans for JSON objects in the text using json.Decoder (which handles trailing content). Also adds diagnostic logging when parsing fails so we can see the actual MCP result text in logs. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Integrate MCP resources protocol (resources/list, resources/read) to discover and load skill guides from the MCP server. Skills are markdown documents at skills/{slug}.md that provide detailed workflow instructions. The skill list is injected into the system prompt so the LLM knows what's available, but skill content is only loaded on-demand via the new get_skill tool when relevant to the user's request. This keeps the context window lean as the skill library grows. - MCP client: add ListSkills, ReadSkill, SkillSummary with TTL caching - Agent: extend MCPToolProvider interface with skill methods - Agent: inject skill summary into system prompt after tool descriptions - Tools: add get_skill native tool (only registered when skills exist) - Executor: add get_skill handler delegating to MCP ReadSkill - Main: pre-warm skill cache at startup alongside tools Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The MCP server returns skill resources with URIs like "skill://vultisig/evm-contract-call.md" but extractSkillSlug was filtering for the "skills/" prefix, discarding all entries. - extractSkillSlug now extracts the last path segment before .md, handling any URI scheme (skill://, skills/, etc.) - ReadSkill now looks up the full URI from the skill cache instead of constructing it, so it works with any URI format Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…rmat OpenRouter's universal format is OpenAI Chat Completions — it translates that to whatever the underlying model needs. Non-Anthropic models were rejecting our Anthropic Messages API format with "Invalid Anthropic Messages API request". - Rewrite client.go types and wire format (ToolCall, AssistantMessage, ToolMessage, ToolChoice with custom MarshalJSON) - URL: /messages → /chat/completions - System prompt: moved from request field to system message - Tool definitions: InputSchema wrapped as function parameters on wire - Response: parsed from choices[0].message with convenience fields - Streaming: OpenAI SSE format (data: chunks, data: [DONE]) - Update agent.go ProcessMessage/ProcessMessageStream for new types - Rename GetAnthropicTools → GetTools in MCP client and interface Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

RaghavSood mentioned this pull request Feb 22, 2026

Add find_token MCP tool support #3

Merged

5 tasks

neavra reviewed Feb 23, 2026

View reviewed changes

RaghavSood mentioned this pull request Feb 23, 2026

Add MCP skill discovery and on-demand loading #6

Merged

7 tasks

RaghavSood force-pushed the feature/mcp-vault-info branch from d4afa11 to edb69f7 Compare February 24, 2026 04:04

RaghavSood and others added 6 commits February 24, 2026 13:25

RaghavSood force-pushed the feature/mcp-vault-info branch from 0ee21c3 to ece7411 Compare February 24, 2026 05:26

RaghavSood merged commit ed20297 into main Feb 24, 2026
1 check passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

MCP server integration + per-conversation vault binding#2

MCP server integration + per-conversation vault binding#2
RaghavSood merged 7 commits intomainfrom
feature/mcp-vault-info

RaghavSood commented Feb 20, 2026

Uh oh!

neavra left a comment

Uh oh!

neavra Feb 23, 2026

Uh oh!

neavra Feb 23, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

RaghavSood commented Feb 20, 2026

Summary

Why this is needed

How it works

MCP client (internal/mcp/client.go)

Vault lifecycle

Configuration

Database

Test plan

Uh oh!

neavra left a comment

Choose a reason for hiding this comment

Uh oh!

neavra Feb 23, 2026

Choose a reason for hiding this comment

Uh oh!

neavra Feb 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

MCP client (`internal/mcp/client.go`)

neavra Feb 23, 2026 •

edited

Loading