MCP Server

Easiest path (UI, free tier): Open Claude Desktop → Customize → Connectors → + → Add custom connector. Paste https://api.algovault.com/mcp?src=docs, then click Add. Enable it per chat from + → Connectors. For a paid-tier key, use the JSON path below.

JSON path: Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "algovault": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.algovault.com/mcp?src=docs",
               "--header", "Authorization:${AUTH_HEADER}",
               "--header", "X-AlgoVault-Track-Token:chan-docs"],
      "env": { "AUTH_HEADER": "Bearer av_live_…" }
    }
  }
}

Set your key in the env block. Free tier: put a free key in the Authorization header for 200 calls a month. Or drop it and the env block to try 20 calls. Keep the X-AlgoVault-Track-Token header either way.

Verify: ask Claude "Get me a trade call for BTC on the 1h timeframe". Tool indicator appears bottom-right of the input box.

Edit ~/.cursor/mcp.json (global, all projects) or .cursor/mcp.json in the project root (per-project, commit-friendly):

{
  "mcpServers": {
    "algovault": {
      "url": "https://api.algovault.com/mcp?src=docs",
      "headers": {
        "Authorization": "Bearer ${env:AV_API_KEY}",
        "X-AlgoVault-Track-Token": "chan-docs"
      }
    }
  }
}

Set AV_API_KEY in your shell. Restart Cursor. The Cursor agent now has AlgoVault tools available while editing strategy code.

Open the Cline panel in VSCode → MCP Servers → Remote Servers tab → Add server. Or edit cline_mcp_settings.json (path varies by OS; access via Configure MCP Servers):

{
  "mcpServers": {
    "algovault": {
      "type": "streamableHttp",
      "url": "https://api.algovault.com/mcp?src=docs",
      "headers": {
        "Authorization": "Bearer ${env:AV_API_KEY}",
        "X-AlgoVault-Track-Token": "chan-docs"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}

type: "streamableHttp" is the modern transport (recommended). The legacy "sse" type still works but is being deprecated upstream.

One-liner (recommended):

claude mcp add --transport http --scope project algovault "https://api.algovault.com/mcp?src=docs" \
  --header "Authorization: Bearer $AV_API_KEY" \
  --header "X-AlgoVault-Track-Token:chan-docs"

This writes a .mcp.json in your repo root which you can commit so every teammate gets the same MCP config:

{
  "mcpServers": {
    "algovault": {
      "type": "http",
      "url": "https://api.algovault.com/mcp?src=docs",
      "headers": {
        "Authorization": "Bearer ${AV_API_KEY}",
        "X-AlgoVault-Track-Token": "chan-docs"
      }
    }
  }
}

Verify: in Claude Code, run /mcp to list connected servers; AlgoVault should appear with its tools.

The Smithery CLI installs and configures the MCP server in your client of choice automatically:

# Pick one — replace <client> with: claude, cursor, cline, claude-code
npx -y @smithery/cli mcp add algovault/crypto-quant-signal-mcp --client <client>

The CLI writes a Smithery-gateway entry into your client's config. It carries no API key, so this path is the free tier. Easiest path if you're new to MCP. Browse the AlgoVault listing at smithery.ai.

For non-MCP integrations (bots, scripts, services), call the JSON-RPC endpoint directly. The transport is stateless, so a single POST of tools/call works: no initialize, no session id. See Testing with raw HTTP / curl for the one-shot call, the two Accept types you must send, and the optional session handshake.

One-shot smoke (free tier, no auth):

curl -sS https://api.algovault.com/health

Returns {"status":"ok","server":"crypto-quant-signal-mcp","version":"<current release>","stripe":true}.

Codex reads MCP servers from ~/.codex/config.toml. Add a table for AlgoVault:

[mcp_servers.algovault]
url = "https://api.algovault.com/mcp?src=docs"
bearer_token_env_var = "AV_API_KEY"

[mcp_servers.algovault.http_headers]
"X-AlgoVault-Track-Token" = "chan-docs"

Set AV_API_KEY in your shell for paid tier; drop bearer_token_env_var for free tier. codex mcp add algovault --url "https://api.algovault.com/mcp?src=docs" adds the free tier from the CLI; the paid tier's bearer_token_env_var is set in config.toml, as above.

IDE extension: open settings, choose MCP servers, add a server, pick Streamable HTTP and paste the same URL.

Verify: ask Codex "Get me a trade call for BTC on the 1h timeframe".

Edit ~/.kimi-code/mcp.json (user level) or .kimi-code/mcp.json (project level). An entry carrying a url and no transport is an HTTP server:

{
  "mcpServers": {
    "algovault": {
      "url": "https://api.algovault.com/mcp?src=docs",
      "bearerTokenEnvVar": "AV_API_KEY",
      "headers": {
        "X-AlgoVault-Track-Token": "chan-docs"
      }
    }
  }
}

Prefer the guided path? Run /mcp-config in the TUI to add, edit or delete servers without touching the JSON.

Verify: ask Kimi "Get me a trade call for BTC on the 1h timeframe".

Open Settings → MCP Servers, then click New MCP Server at the top right. Choose HTTP as the type and enter:

https://api.algovault.com/mcp?src=docs

For paid tier, expand Headers (optional) and add Authorization: Bearer av_live_…. Free tier needs no header.

ZCode also accepts a pasted config block under Full configuration, in either the {"mcpServers": {…}} or the bare {"server-name": {…}} shape.

Verify: ask ZCode "Get me a trade call for BTC on the 1h timeframe".

One step: patch the profile. There is no plugin to install — the dsh CLI ships @deepseek-ai/dsh-mcp-client in its own dependency closure, and the bridge’s README says one entry per server is the entire setup.

The bundles are a red herring: base, headless and web-app declare zero MCP dependencies, but a bare plugin name resolves through the profile’s Node parent walk to $DSH_HOME/profiles/node_modules, which the CLI closure feeds. Nothing is enabled by default because DSH treats each server as trusted code outside the sandbox — the entry below is the opt-in.

Edit ~/.dsh/profiles/<name>/cordis.patch.yml, or ~/.dsh/cordis.patch.yml for every profile:

- insert:
    - id: mcp-algovault
      name: '@deepseek-ai/dsh-mcp-client'
      config:
        serverName: algovault
        transport: streamable-http
        url: https://api.algovault.com/mcp?src=deepseek_harness
        headers:
          X-AlgoVault-Track-Token: int-deepseek-harness

The - insert: wrapper is required. Without it the entry is an id-targeted override, and it is skipped with a warning. The bridge’s own README shows the unwrapped form because it documents the plugin config shape, not a cordis.patch.yml edit.

Paid tier adds one more line to that headers block: Authorization: Bearer av_live_…. Without a key you get 20 calls a month; a free key gives 200.

The tools arrive server-qualified — mcp__algovault__get_trade_call, mcp__algovault__scan_trade_calls, and the rest.

DSH bridges tools only. MCP resources and prompts are deferred upstream, so read the track record at algovault.com/track-record instead.

Verify: ask dsh "Get me a trade call for BTC on the 1h timeframe".

Verified against dsh 0.1.1-rc.2 on 2026-08-29. DSH ships prereleases only; expect compatibility-breaking changes.

Z.ai dials the MCP server itself, so there is nothing to install locally. Declare AlgoVault as a tool on the request:

{
  "model": "glm-4.6",
  "messages": [{"role": "user", "content": "Trade call for BTC on the 1h timeframe"}],
  "tools": [{
    "type": "mcp",
    "mcp": {
      "server_label": "algovault",
      "server_url": "https://api.algovault.com/mcp?src=docs",
      "headers": {"X-AlgoVault-Track-Token": "chan-docs"}
    }
  }]
}

server_label is required. transport_type is optional and already defaults to streamable-http, so it is omitted above. Add allowed_tools to narrow the tool set.

DeepSeek’s own harness connects to AlgoVault directly — see the DeepSeek Harness tutorial. This row is the other path: keep the harness you already run and swap the model behind it. The DeepSeek API itself still exposes no MCP parameter, so the harness carries the tools.

export ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic"
export ANTHROPIC_AUTH_TOKEN="$DEEPSEEK_API_KEY"

claude mcp add --transport http --scope project algovault \
  "https://api.algovault.com/mcp?src=docs" \
  --header "X-AlgoVault-Track-Token:chan-docs"

Claude Code then runs against DeepSeek while AlgoVault stays connected exactly as it would otherwise. Verdicts are unchanged: they are computed on our side and handed back as JSON.

Harnesses that speak the OpenAI protocol instead use the plain https://api.deepseek.com base, not the /anthropic one shown here.