Skip to main content
Glama

mcp-gtags-server

Stop letting your AI agent grep. Give it an index.

PyPI CI Kernel eval Python 3.10+ License: MIT Docs MCP Powered by GNU Global Install in Cursor

mcp-name: io.github.harshithsunku/mcp-gtags-server

# If you don't have uv yet (one-time, no sudo):
curl -LsSf https://astral.sh/uv/install.sh | sh
{ "mcpServers": { "gtags": { "command": "uvx", "args": ["mcp-gtags-server"] } } }

One user-level config entry, no sudo, no pre-installed anything — the whole toolchain installs itself into user space on first use, and every repo you open is served automatically.

📖 Full documentation — install for every client, the tool reference, output format, kernel features, configuration and troubleshooting.

Every AI coding agent — Claude Code, Cursor, Codex, you name it — answers "where is this function defined?" the same way: grep the entire tree. On a million-line C/C++ codebase that's a full scan per question, and the output is a firehose: every comment, string literal, and unrelated match, dumped straight into the model's context window.

mcp-gtags-server replaces those scans with indexed lookups powered by GNU Global (gtags) — the same tags engine kernel and systems developers have trusted for decades — exposed to agents over the Model Context Protocol. Built for the codebases LSP-based tools can't handle: kernel-scale C/C++, trees that don't currently compile, machines you can't sudo on.

  • ~100× faster per query — milliseconds instead of seconds, at any codebase size

  • Radically less noise — the definition, not 7,873 lines of matches

  • Speaks kernel — #ifdef guard stacks with .config filtering, macro-generated symbols (sys_read → its SYSCALL_DEFINE3 site) that no other tagging tool resolves, and definitions the parser misses recovered from their EXPORT_SYMBOL site via ctags

  • Zero index management — first query builds the index, every query auto-refreshes it

  • Correctness measured in CI — a 64-case golden eval covering all 8 tools against a pinned kernel: 100% recall, 100% precision@1 (docs/capability.md)

  • Works everywhere MCP does — Claude Code, Claude Desktop, Cursor, any MCP client

The numbers (real Linux kernel, not a toy)

Measured on a full Linux kernel checkout — 65,163 C/C++ files, 37.1 million lines — warm page cache:

Question an agent asks

grep -rn

gtags (this server)

Context consumed

Where is tcp_v4_rcv defined?

1.40 s

0.01 s

8 lines → 1 line

Where is kmalloc defined?

1.62 s

0.01 s

7,873 lines → 5 lines

Who references kmalloc?

1.62 s

0.10 s

7,873 noisy lines → 2,744 real sites (or a ranked per-file summary)

Show me tcp_v4_rcv's implementation

read a 3,500-line file

get_symbol_body

exactly the 271-line function

Who calls ext4_mark_inode_dirty?

245 raw match lines

find_callers

62 deduped caller functions, with counts

Where is sys_read really defined?

no answer — the name is macro-generated

0.03 s

fs/read_write.c SYSCALL_DEFINE3(read, ...), flagged resolved_via

Where is mutex_lock defined?

24,774 noisy match lines

0.2 s

kernel/locking/mutex.c:314 — recovered via EXPORT_SYMBOL + ctags after gtags' parser derails on it

Does ksys_read ever reach rw_verify_area?

N rounds of grep + reading

0.6 s

the shortest call chain, with every call site's file:line

Every mutex_lock call in fs/ext4?

22,905 tree-wide match lines

0.01 s

find_references(path_prefix="fs/ext4"): the 18 real sites

One-time index build: 66 s for the whole kernel. Incremental refresh after edits: well under a second. Reproduce it yourself with scripts/benchmark.sh:

./scripts/benchmark.sh /path/to/linux tcp_v4_rcv kmalloc ext4_readdir

The speed is nice. The real win is precision: an agent that gets 5 exact lines instead of 7,873 noisy ones keeps its context window for actual reasoning.

And the answers are measured, not assumed: CI runs a 64-case golden eval against pinned kernel v6.16 on every push — currently 100% recall, 100% precision@1 across definitions, macro resolution, export recovery, references, callers, callees, definition bodies, #ifdef guards, and reachability, covering all 8 tools. The full methodology, numbers, and honest limitations live in docs/capability.md.

Related MCP server: lsp-mcp-server

Quick start (60 seconds)

Option A — one config entry (recommended)

Add the server once, at user level, and you're done — no installer, no pre-installed gtags, no per-repo setup. The only prerequisite is uv (or use Option B, which installs it for you).

# Claude Code (once per device, all repos):
claude mcp add --scope user gtags -- uvx mcp-gtags-server
// Cursor (~/.cursor/mcp.json), or any MCP client's global settings:
{
  "mcpServers": {
    "gtags": { "command": "uvx", "args": ["mcp-gtags-server"] }
  }
}

Or click the Cursor one-click install badge at the top of this page.

On the very first tool call the server bootstraps everything else by itself: GNU Global (prebuilt user-space binaries — no compiler, no sudo), universal-ctags, and Pygments, all into ~/.gtags-mcp. While that one-time install runs (a few seconds on most platforms), tool calls return a "toolchain is being installed — retry shortly" status instead of failing. Opt out with --no-auto-setup or GTAGS_MCP_AUTO_SETUP=0.

The prebuilt Linux binaries run on any distro with glibc ≥ 2.28 (RHEL/Rocky 8+, Ubuntu 18.10+, Debian 10+). On older hosts — or if a downloaded binary fails its post-install execution check — setup automatically compiles GNU Global from source instead, which only needs make and a C compiler.

Option B — install the plugin (server + skill + slash commands)

The plugin bundles this server, a "C/C++ code navigation" skill that tells the agent which tool answers which question, and two slash commands (/gtags:impact, /gtags:explain). Installing it registers the MCP server for you — there is no config file to edit.

Client

Install

Claude Code

/plugin marketplace add harshithsunku/mcp-gtags-server then /plugin install mcp-gtags-server@mcp-gtags-server

Codex

codex plugin marketplace add harshithsunku/mcp-gtags-server, then install it from the Plugins Directory in the ChatGPT desktop app

Cursor

Loads Agent Plugins 1.0.0, but installs only from the Cursor Marketplace or a team marketplace — until this plugin is listed there, use the one-click badge or Option A

Install one way, not both. A plugin and a manual entry run two servers, and the agent sees every tool twice. Switching to the plugin? Remove the manual entry first (claude mcp remove gtags).

Option C — one-line installer (shared background server)

One command. No sudo. Works everywhere — restricted corporate machines, containers, build servers:

curl -fsSL https://raw.githubusercontent.com/harshithsunku/mcp-gtags-server/main/scripts/install.sh | bash

Everything lands in your home directory — the server (via uv), GNU Global, universal-ctags, and Pygments (in ~/.gtags-mcp). When it finishes, a shared background HTTP server is running that every client and IDE window on the machine can point at (http://127.0.0.1:8383/mcp), and the exact client configuration is printed to your console.

Re-run the same command any time:

  • Up to date? → "Already installed and up to date — nothing to install", and the config is printed again.

  • New release on GitHub/PyPI? → the package updates, an outdated gtags toolchain is wiped and reinstalled automatically, and the background server restarts on the new version.

Slash commands (MCP prompts)

Available with any install, in clients that support MCP prompts (Claude Code, Cursor). They cost no tool-schema context:

Command

What it runs

/gtags:impact [git_ref]

git diff → the changed functions → find_callers on each → callers ranked by risk

/gtags:explain <symbol>

definition (+ #ifdef variants) → body → callers → a short explanation

One config, many repos

However you install it, one user-level entry serves every repo you open — 20 repos need zero extra installs and zero extra config. Each tool call resolves its project root down this ladder:

  1. project_root argument on the tool call (agents pass this to target any tree)

  2. --root flag / GTAGS_MCP_ROOT env var

  3. root in a config file (note: pinning a root here defeats multi-repo)

  4. The client's workspace roots (MCP roots protocol) — IDEs that advertise their open folders get the right repo automatically, even on the shared HTTP server; with several folders open, agents are asked to pass project_root. Roots exist only for clients on protocol revisions up to 2025-11-25: the 2026-07-28 revision deprecated them, so those clients skip this step (stdio servers still resolve through step 5; on the shared HTTP server, agents pass project_root)

  5. Walk up from the server's working directory to the nearest .git/GTAGS — this is why stdio servers spawned by Claude Code/Cursor inside a repo just work

Step 5 can be switched off with --no-cwd-fallback / GTAGS_MCP_CWD_FALLBACK=0 (the plugin sets it): the tools then ask for project_root instead of indexing whatever directory the server happens to run in. Plugin clients launch the server from the plugin's own folder, which is never your project.

That's it. No indexing step, no configuration. Ask your agent "who calls tcp_v4_rcv?" — the first query in any repo builds that repo's index automatically, and every query after that is answered in milliseconds. Run mcp-gtags-server doctor any time to see what the server detects, or mcp-gtags-server config to re-print the client configuration.

The installer runs mcp-gtags-server --transport http --host 127.0.0.1 --port 8383 in the background (pid: ~/.gtags-mcp/server.pid, log: ~/.gtags-mcp/server.log). Environment overrides for the installer:

Variable

Default

Meaning

GTAGS_MCP_PORT

8383

HTTP port

GTAGS_MCP_HOST

127.0.0.1

Bind address — set 0.0.0.0 to reach the server from other devices at http://<machine-ip>:8383/mcp

GTAGS_MCP_NO_SERVER

unset

1 = don't start a background server

Security note: the HTTP endpoint is unauthenticated. It binds localhost by default; only bind 0.0.0.0 on networks you trust.

# 1. GNU Global — EITHER user-space (no sudo):
mcp-gtags-server setup
#    OR a system package:
sudo apt install global      # Debian/Ubuntu
sudo dnf install global      # Fedora
brew install global          # macOS

# 2. The server:
uv tool install mcp-gtags-server        # or: pip install mcp-gtags-server

The server finds binaries in this order: --bin-dir/GTAGS_MCP_BIN_DIR/config bin_dir → ~/.gtags-mcp/bin → PATH → ~/.local/bin.

Easiest: download mcp-gtags-server.mcpb from the latest release, drag it into Claude Desktop's Settings → Extensions, and pick your project folder in the setup screen. The gtags toolchain installs itself on first use.

Or add to claude_desktop_config.json manually (pin the project since Desktop doesn't launch in your repo):

{
  "mcpServers": {
    "gtags": {
      "command": "uvx",
      "args": ["mcp-gtags-server", "--root", "/absolute/path/to/your/project"]
    }
  }
}

By default the server uses the client's advertised workspace root (MCP roots protocol) or auto-detects the project root by walking up from its working directory to the nearest .git or existing GTAGS — so queries from anywhere inside a monorepo resolve to the repo root. Override with --root /path, the GTAGS_MCP_ROOT env var, or root in a config file — or pass project_root on any individual tool call to query a different tree. (See "One config, many repos" for the full resolution order.)

Every setting can also live in a TOML file, so teams share defaults through the repo (like .editorconfig):

  • Project: .gtags-mcp.toml at the project root

  • User: ~/.config/gtags-mcp/config.toml

# .gtags-mcp.toml
label = "native-pygments"     # force a GTAGSLABEL parser label
bin_dir = "/opt/tools/bin"    # extra directory searched for gtags/global/ctags
skip_globs = ["*.gen.c"]      # never index paths/basenames matching these globs
respect_gitignore = true      # default: index only what `git ls-files` reports
enrich = true                 # default: ctags kind/signature/scope on results
guards = true                 # default: #ifdef guard stacks on results
macro_resolve = true          # default: resolve macro-generated symbols (sys_*, ...)
# root = "/abs/path"          # default project root (user config)

Precedence: tool-call argument > CLI flag > environment variable > project config > user config > built-in default.

Tools

Eight tools, all read-only except update_index. Each takes project_root and format ("text" default, "json"); list-shaped tools also take limit/offset. Full parameter reference: docs → Tools.

  • find_definition — Go to definition: every definition with its #ifdef guard stack and ctags kind/signature, plus a usage summary (reference/file counts, hottest files, EXPORT_SYMBOL* status). Resolves macro-generated names (sys_read → SYSCALL_DEFINE3).

  • find_references — Every usage site with its guard stack; above 200 references it returns a per-file distribution, and path_prefix narrows it to one directory.

  • get_symbol_body — The source of one definition (function, struct, or multi-line macro), not the whole file.

  • find_callers — Who calls this function: references mapped to their enclosing functions, deduplicated, with call counts and call lines.

  • find_callees — What this function calls: call sites taken from its body and checked against the index, split into in-tree (with locations) and external.

  • reachability — The shortest static call chain from function A to function B, with the file:line of every hop, or "no static path".

  • list_file_symbols — Every symbol a file defines (its API surface).

  • update_index — Synchronous index refresh after edits; full=true rebuilds from scratch. The only tool that writes.

Symbol-level tools — the noise killers

Give the agent the symbol, not the file.

Tool

What the agent gets

get_symbol_body

Just the source of a definition. The 271-line tcp_v4_rcv function — not the 3,500-line file it lives in. Handles functions, structs, and multi-line macros.

find_callers

The call graph, deduplicated. Every reference mapped to its enclosing function with call counts: 245 raw lines for ext4_mark_inode_dirty collapse to 62 callers. Iterate it to walk the caller graph as deep as you need.

find_callees

The outgoing call graph. What does this function call? Body-extracted call sites, each verified against the index, split into in-tree (with locations) and external.

reachability

"Can this function end up in that one?" — BFS over the caller graph returns the shortest call chain from A to B with the file:line of every call site (ksys_read → vfs_read → rw_verify_area), or an honest "no static path" that names the function-pointer caveat. One call instead of a dozen find_callers rounds.

Core lookups

Tool

What it does

Underlying command

find_definition

Where is this defined, and what is it? Definitions with #ifdef guards and ctags metadata, plus a usage summary (reference/file counts, hottest files, EXPORT_SYMBOL* status). Falls back to macro-family resolution (sys_*, trace_*, DEFINE_* names) and suggests prefix matches on a miss

global -x + macro resolution

find_references

Every usage site, each with its guard stack. Above 200 references it returns a per-file distribution instead of a wall of lines (kmalloc's 2,744 references become one screen); path_prefix drills into one directory, group_by forces either view. Falls back to symbol-usage records (libc calls, some variables), flagged fallback

global -rx -S <dir>, fallback global -sx

list_file_symbols

A file's API surface — every symbol it defines

global -fx

update_index

Synchronous freshness barrier after edits; full=true rebuilds from scratch (rarely needed — indexing is automatic)

gtags -i / gtags

Every query tool supports limit/offset pagination, long-line truncation, and (where it makes sense) case_insensitive — output is engineered to never flood a context window.

Every tool also declares MCP tool annotations: all are readOnlyHint: true except update_index, and none reach outside your machine (openWorldHint: false). Clients use these hints to auto-approve and parallelize read-only calls. Each response carries the envelope exactly once, as text content, with no duplicate structuredContent copy.

The flow that saves your context window

1. find_definition("kmalloc")                → definitions + #ifdef variants + usage spread (3 lines)
2. find_callers("ext4_mark_inode_dirty")     → deduped callers, each with its call line
3. get_symbol_body("tcp_v4_rcv")             → read the ONE function that matters
4. find_callees("tcp_v4_rcv")                → what it depends on, with locations
5. reachability("ksys_read", "rw_verify_area") → the call chain, one line per hop
6. find_references("mutex_lock", path_prefix="fs/ext4") → hot symbol, one subsystem

A few hundred lines of context total — versus tens of thousands for the grep-and-read-files equivalent.

Output and kernel smarts

Output: compact text by default, JSON on request

Since v2.0.0 every tool answers with grep-shaped text — the format agents read best, and 2–3× cheaper in tokens than the same answer as JSON:

kmap: 4 definitions under 3 #if variants · 261 refs in 65 files (top: tools/perf/util/machine.c 26, …)
include/linux/highmem-internal.h:40: static inline void *kmap(struct page *page)  [function; #if CONFIG_HIGHMEM]
include/linux/highmem-internal.h:170: static inline void *kmap(struct page *page)  [function; #if !CONFIG_HIGHMEM]
next: get_symbol_body, find_callers, find_references

path:line: source rows carry a tag with what the symbol is and the #if/#ifdef stack it lives under. Paginated results end with a footer like [1-100 of 5290 · offset=100 for more].

Against v1.x's JSON default the same seven kernel questions cost 20% fewer tokens overall — up to 68% on individual lookups. find_callers is the one call that got bigger (about 9%), on purpose: it now includes each call site's source line, which is exactly what agents used to spend an extra grep to see.

Pass format="json" for the machine-readable envelope — the same data, rendered from one source of truth:

{
  "tool": "find_definition",
  "root": "/abs/project/root",
  "results": [
    {"symbol": "kmap", "path": "include/linux/highmem-internal.h", "line": 40,
     "col": 22, "kind": "function", "typeref": "void *", "scope": null,
     "signature": "(struct page * page)", "guard": ["CONFIG_HIGHMEM"],
     "snippet": "static inline void *kmap(struct page *page)"}
  ],
  "total": 2, "offset": 0, "truncated": false,
  "definition_count": 2, "guard_variants": 2, "reference_count": 261,
  "file_count": 65, "top_files": [{"path": "…", "count": 26}], "exported": null,
  "next_tools": ["get_symbol_body", "find_callers", "find_references"],
  "warning": null
}
  • Symbol locations always use the stable record schema {symbol, path, line, col, kind, typeref, scope, signature, guard, snippet} with repo-relative paths. Keys are only ever added, never renamed or removed within a major version.

  • kind / typeref / scope / signature say what a symbol is (since v0.8.1): function vs. macro vs. struct vs. typedef vs. enum constant, its return/target type, its enclosing scope (enum:color, struct:item), and its parameter list — extracted per file by universal-ctags with no build and no compile database, cached, and filled on definition-shaped results (find_definition, list_file_symbols). When universal-ctags isn't available the fields are simply null; disable explicitly with --no-enrich, GTAGS_MCP_ENRICH=0, or enrich = false in .gtags-mcp.toml.

  • guard says when a symbol exists (since v0.9.0): the enclosing #if/#ifdef stack, outermost first ([] = unconditional, null = scanning disabled or file unreadable). See the next section — this is the headline feature.

  • resolved_via says how a symbol was found when it took macro-family resolution rather than a literal index match ("macro:SYSCALL_DEFINE", "fuzzy:vfs_read") — see Macro-generated symbols.

  • next_tools tells the agent the highest-value follow-up call for what was (or wasn't) found.

  • total/offset/truncated drive pagination; failures keep the envelope with an error field and are flagged isError on the protocol so the agent can self-correct.

  • Composite tools return tool-shaped results (e.g. find_callees {in_tree, external}, reachability a hop chain) inside the same envelope.

  • Every tool declares MCP tool annotations: all are readOnlyHint: true except update_index, and none reach outside your machine (openWorldHint: false) — clients use these to auto-approve and parallelize calls.

Upgrading from v1.x? Pass format="json" explicitly wherever you parsed the old default. symbol_info is now part of find_definition, summarize_references is find_references (it groups by file automatically), and blast_radius is the /gtags:impact prompt — git diff, then find_callers on each changed function.

#ifdef-aware: know which definition your config actually compiles

Kernel and firmware code defines the same symbol multiple times and lets the build configuration pick one. Every other no-build tool returns a flat, unexplained list — an agent happily reads the no-op stub of kmap and reasons its way to a wrong answer. This server reads the preprocessor conditionals (pure scanning — still no build, no compile_commands.json):

Symbol: kmap
  2 definitions under 2 distinct guards:
  [CONFIG_HIGHMEM] defined at include/linux/highmem-internal.h:40 — function kmap(struct page * page) -> void *
  [!CONFIG_HIGHMEM] defined at include/linux/highmem-internal.h:170 — function kmap(struct page * page) -> void *

Pass active_config — a kernel .config path or a macro list like "CONFIG_SMP,BITS_PER_LONG=64,!CONFIG_DEBUG" — to find_definition, or find_references, and definitions whose guard stack is definitely false under it are dropped (the envelope reports the count as config_filtered). Filtering is deliberately conservative: a .config is a closed world for CONFIG_* macros (kbuild semantics, including =m → CONFIG_X_MODULE and IS_ENABLED/IS_BUILTIN/IS_MODULE), but anything unknown (__ASSEMBLY__, ARCH_HAS_*, arithmetic it can't decide) never drops a result.

The details are handled so the output stays clean: classic include guards (#ifndef FOO_H) are detected and suppressed, #elif chains compose into explicit conditions (!CONFIG_X86_64 && CONFIG_X86_32), comments on directives are ignored (they lie), and broken/partial files never fail a query. Disable with --no-guards, GTAGS_MCP_GUARDS=0, or guards = false in .gtags-mcp.toml.

Macro-generated symbols resolve too (sys_read → SYSCALL_DEFINE3)

The best-known gap of every tagging tool on the kernel: sys_read, trace_sched_switch, and css_set_lock have no literal definition anywhere — they are minted by token-pasting macros, and a plain lookup comes back empty (or worse, returns a same-named test helper). This server resolves them from the index alone, no preprocessor and no build:

find_definition("sys_read")        → fs/read_write.c:723  SYSCALL_DEFINE3(read, ...)   resolved_via: "macro:SYSCALL_DEFINE"
find_definition("__x64_sys_openat")→ fs/open.c:1385       SYSCALL_DEFINE4(openat, ...) (arch entry-point wrappers map back)
find_definition("trace_sched_switch") → include/trace/events/sched.h:220  TRACE_EVENT(sched_switch, ...)
find_definition("css_set_lock")    → kernel/cgroup/cgroup.c:82  DEFINE_SPINLOCK(css_set_lock);

Covered families: SYSCALL_DEFINE0..6 / COMPAT_SYSCALL_DEFINE* (incl. __x64_/__ia32_/__arm64_/__se_/__do_ wrapper spellings), tracepoints (TRACE_EVENT, DEFINE_EVENT, DECLARE_TRACE, ...), and the bare-name definers (DEFINE_SPINLOCK, DEFINE_MUTEX, DEFINE_PER_CPU*, DECLARE_BITMAP, module_param*, any DEFINE_/DECLARE_-shaped macro) — plus a last-resort fuzzy tier that tries underscore-variant spellings. Resolved results are flagged with resolved_via in the envelope and ranked ahead of same-named textual shadows; DEFINE_* sites rank above their DECLARE_* counterparts. find_definition additionally reports the EXPORT_SYMBOL / EXPORT_SYMBOL_GPL variant a kernel symbol is exported with, in an exported field. Costs nothing when a symbol resolves normally (only family-shaped names like sys_*/trace_* get the extra indexed lookups); disable with --no-macro-resolve, GTAGS_MCP_MACRO_RESOLVE=0, or macro_resolve = false.

What gets indexed (junk stays out)

Indexing feeds gtags an explicit file list instead of letting it walk the tree:

  • In a git repository the list comes from git ls-files — .gitignore is respected exactly, so build output, vendored blobs, and generated files never pollute the index. Disable with respect_gitignore = false in .gtags-mcp.toml.

  • Outside git, the tree is walked minus well-known junk directories (.git, node_modules, build, dist, .venv, ...).

  • skip_globs in .gtags-mcp.toml drops anything else you never want indexed.

Incremental refreshes recollect the list, so newly ignored files drop out of the index and new files appear — automatically.

Multi-language projects (C + Python + more)

Real projects mix languages — a C core with Python tooling, JS frontends, Go services. The server handles this automatically:

  • Native languages (C, C++, Java, PHP, Yacc, assembly) use GNU Global's fast built-in parser.

  • Everything else (Python, Go, Rust, JavaScript, TypeScript, Ruby, ... ~150 languages) is indexed through Global's ctags + Pygments plugin parsers — same index, same tools, same queries.

The one-line installer (and mcp-gtags-server setup) enables this automatically — it installs universal-ctags and Pygments into user space, and the server switches to the native-pygments parser label on its own. Prefer system packages? Those work too:

sudo apt install exuberant-ctags python3-pygments   # Debian/Ubuntu
sudo dnf install ctags python3-pygments             # Fedora
brew install ctags && pip install pygments          # macOS

Now find_definition("py_util"), get_symbol_body (indentation-aware for Python), find_callees, find_callers — all work across every language in the tree, in one index.

Force a specific parser label with --label, GTAGS_MCP_LABEL, or label in .gtags-mcp.toml (e.g. default for native-only, pygments for plugin-everything).

Honest caveats: for plugin-parsed languages, definitions are as accurate as ctags, but references are token-based — every occurrence of the name counts, without C-grade semantic reference tracking or local-scope awareness. For C/C++ nothing changes: the native parser still does that part.

How it works

Architecture

One long-lived server process per IDE window (stdio) — or one shared HTTP server for the whole machine. Everything heavy lives on disk and is shared: the toolchain installs itself once per machine, the index builds itself once per repo.

flowchart LR
    subgraph clients["MCP clients"]
        CC["Claude Code / Cursor / VS Code"]
        CD["Claude Desktop (.mcpb)"]
    end

    subgraph proc["mcp-gtags-server — spawned once per window, lives for the session"]
        T["11 navigation tools"]
        RR["root resolution<br/>project_root → env → config → client roots → cwd"]
    end

    subgraph disk["your machine, user space (no sudo)"]
        TC["~/.gtags-mcp<br/>global · gtags · ctags · Pygments<br/>self-installs on first use"]
        IX["&lt;repo&gt;/.gtags-mcp/GTAGS<br/>one index per repo, auto-built"]
    end

    CC -- "stdio (uvx)" --> T
    CD -- "stdio" --> T
    T --> RR
    RR -- "runs global (ms, indexed)" --> TC
    TC -- "B-tree lookup" --> IX

A tool call, end to end

The binary is not re-executed per call — the server process stays alive; each call is one JSON-RPC message plus one millisecond-scale global subprocess:

sequenceDiagram
    participant A as Agent
    participant S as server (long-lived process)
    participant G as global (GTAGS index)

    A->>S: find_definition("tcp_v4_rcv")
    Note over S: first call on this machine?<br/>toolchain installs itself in the background,<br/>calls answer "installing — retry shortly"
    Note over S: first query in this repo?<br/>index auto-builds once (~66 s for the kernel)
    S->>G: global -dx tcp_v4_rcv
    G-->>S: net/ipv4/tcp_ipv4.c:2067 (milliseconds)
    S-->>A: narrow JSON answer — no grep firehose
  • First query on a tree? The index is built automatically (the only operation that ever blocks — and only once).

  • Where do the index files go? Into a single .gtags-mcp/ folder at the project root — never loose files next to your code. The folder ships its own .gitignore, so git status stays clean without touching yours. A pre-existing root-level GTAGS (from older versions, or your own gtags runs) keeps being used as-is. mcp-gtags-server doctor shows the location.

  • Files changed? A debounced incremental refresh runs in the background: queries always answer instantly from the current index while gtags -i catches up behind the scenes. Measured on the kernel: queries return in 0.02s while the 25s freshness check runs invisibly. Staleness is bounded by the debounce window; call update_index for a synchronous, guaranteed-fresh barrier right after edits.

  • Huge result? Pagination footers tell the agent exactly how to fetch the next page — or the tool itself suggests a narrower one (find_callers on a symbol used in 500+ files points to find_references, which groups by file).

FAQ

Why gtags instead of a language server (LSP)? LSP servers give richer semantics but need a working build configuration, per-editor setup, and serious warm-up time on large trees. gtags indexes 37M lines in about a minute with zero configuration, handles the kernel-scale codebases LSPs choke on, and its fuzzy parser doesn't care whether the code currently compiles. For C/C++ navigation questions — definition, references, callers — it's the pragmatic sweet spot. (Wrapping clangd for the compile-DB case was considered and deliberately rejected: users who have a working compile_commands.json already have clangd and its ecosystem — see ROADMAP.md.)

What languages? C, C++, Yacc, Java, PHP, and assembly natively — plus Python, Go, Rust, JS/TS, Ruby, and ~150 others via the ctags/Pygments plugin parsers (see Multi-language projects).

Does the agent have to manage the index? No. That's the point. Build-on-first-query, background refresh with adaptive debounce, zero blocking — queries never wait for index maintenance. The explicit update_index tool exists only as an escape hatch: a synchronous freshness barrier after edits, and a from-scratch rebuild with full=true.

Where does the index live? Can I delete it? In .gtags-mcp/ at the project root (self-gitignored). Delete it freely any time — the next query rebuilds it from scratch.

Will it fight my agent's built-in tools? The tool descriptions are written to steer the model: they say when to use indexed lookups instead of grep. In practice agents pick the faster, narrower tool naturally.

ModuleNotFoundError: No module named 'mcp.server.fastmcp' on startup? Releases up to v1.4.2 declared an uncapped mcp dependency, so installs made after the MCP Python SDK 2.0 release resolve an SDK they can't import. v1.4.3 fixes it by capping the SDK below 2.0, and v1.5.0+ runs natively on SDK 2.x (mcp>=2.2,<3), speaking every protocol revision from 2024-11-05 through 2026-07-28. Pick up the fix with uvx --refresh mcp-gtags-server --help (or uvx mcp-gtags-server@latest), or re-run the installer. If you must stay on an older release, pin the SDK yourself: uvx --with 'mcp<2' mcp-gtags-server@1.4.2.

Development

git clone https://github.com/harshithsunku/mcp-gtags-server
cd mcp-gtags-server
uv run --extra dev pytest       # 325 tests; e2e tests auto-skip if GNU Global is absent
npx @modelcontextprotocol/inspector mcp-gtags-server    # poke at it interactively

Tests build a real C project in a temp dir and exercise auto-indexing, auto-refresh, caller mapping, body extraction, pagination, user-space binary discovery, and config layering end-to-end.

Correctness is also measured, not assumed: mcp-gtags-server eval --golden evals/golden.jsonl --root <kernel-tree> runs a 64-case golden set covering all 8 tools (definitions, macro resolution, references, callers, callees, bodies, guards, reachability, maintenance) against a real kernel and prints recall / precision@1 — CI does this weekly against a pinned tag. See docs/capability.md for the current numbers.

Release flow: bump version in pyproject.toml, tag vX.Y.Z, push — CI publishes to PyPI and users pick the update up on their next installer re-run. Prebuilt GNU Global binaries are rebuilt by tagging global-v<version> (or gh workflow run release-binaries.yml -f version=<version> to replace the assets of an existing release in place); Linux builds run inside manylinux_2_28 containers so they work on any glibc ≥ 2.28 host, enforced by a CI symbol-ceiling check.

Roadmap

See ROADMAP.md — structured JSON output landed in v0.8.0, ctags metadata enrichment (kind/signature/scope) in v0.8.1, #ifdef/config-guard awareness (the headline capability for kernel and firmware trees) in v0.9.0, then, all shipped in v1.0.0: macro-family symbol resolution (sys_read → its SYSCALL_DEFINE3 site), the agent workflow tools (reachability, diff-impact analysis), automatic recovery from corrupted index databases, and the correctness eval harness — a 64-case golden set against pinned kernel v6.16 scoring 100% recall / 100% precision@1 in CI, with the measured writeup in docs/capability.md. Every technical milestone is done; what remains is distribution (MCP registry, directories, the writeup post).

Contributions welcome — open an issue or PR.

License

MIT © Harshith Sunku

Available Tools

8 tools
find_calleesFind calleesA
Read-only

What does this function call? Callees / outgoing calls of a function.

Shows a function's dependencies without reading any file: call sites are detected in its body and verified against the index, split into in-tree functions (with locations) and external/unresolved names. Macro-generated and parser-missed (EXPORT_SYMBOL-recovered) definitions resolve too, flagged resolved_via. JSON results: {in_tree: [{symbol, path, line}], external: [names]}.

Args: symbol: Exact name of the function to analyze.

Inputs: symbol (required), format, project_root.

find_callersFind callersA
Read-only

Who calls this function? Callers / call hierarchy / incoming calls, deduplicated per calling function with call counts.

Each reference is mapped to its enclosing function and shown with the source line of its first call site, so there is nothing to re-grep. The highest signal-to-noise "who uses this?" view for impact analysis; iterate it to walk the caller graph upward. JSON results: {caller, path, sites, call} items (call = source of the first site).

Args: symbol: Exact symbol name whose callers you want.

Inputs: symbol (required), limit, format, offset, +1 more.

find_definitionFind definitionA
Read-only

Go to definition: where a C/C++ symbol (function, struct, macro, typedef, enum) is defined, with a usage summary — the best first query.

Each definition carries its #if/#ifdef guard stack (several guarded definitions = a config choice) and ctags kind/signature. The summary gives reference and file counts, the hottest files and the EXPORT_SYMBOL* variant. Macro-generated symbols resolve ("sys_read" -> SYSCALL_DEFINE3(read, ...), DEFINE_SPINLOCK names), and definitions the index parser missed are recovered from their EXPORT_SYMBOL* site — both flagged resolved_via. A miss suggests similarly named symbols.

JSON: results are {symbol, path, line, col, kind, typeref, scope, signature, guard, snippet} records; the envelope adds definition_count, guard_variants, reference_count, file_count, top_files, exported.

Args: symbol: Exact symbol name, e.g. "tcp_v4_rcv". case_insensitive: Match ignoring case (skips the usage summary). active_config: Kernel .config path or macro list like "CONFIG_SMP,BITS_PER_LONG=64,!CONFIG_DEBUG"; drops definitions whose guard stack is definitely false under it (count reported as config_filtered). Unknown macros never drop anything.

Inputs: symbol (required), limit, format, offset, +3 more.

find_referencesFind referencesA
Read-only

Find all references / usages of a C/C++ symbol — every call and use site.

Only real reference sites from the index, each with its #if/#ifdef guard stack. Very widely used symbols (more than 200 sites, e.g. kmalloc) come back grouped by file with counts — see where usage concentrates, then narrow with path_prefix. Symbols with no in-tree definition (libc calls, some variables) work too, flagged "fallback": "symbol_usages".

Args: symbol: Exact symbol name. case_insensitive: Match ignoring case. active_config: Kernel .config path or macro list; drops references whose guard stack is definitely false under it (config_filtered). group_by: "auto" (default: per-file counts above 200 references), "line" (always individual sites) or "file" (always per-file counts). path_prefix: Only references under this directory, e.g. "fs/ext4".

Inputs: symbol (required), limit, format, offset, +5 more.

get_symbol_bodyGet symbol bodyA
Read-only

Read a symbol's source: the full body of a function, struct or macro definition, without reading the whole file.

Extracts only the definition's lines, so a one-screen function never costs a 5000-line file read. Macro-generated and parser-missed (EXPORT_SYMBOL-recovered) definitions resolve too, flagged resolved_via. JSON results: {path, line, body} items.

Args: symbol: Exact symbol name. max_definitions: Return at most this many bodies when the symbol is multiply defined (default 3).

Inputs: symbol (required), format, project_root, max_definitions.

list_file_symbolsList file symbolsA
Read-only

Outline of one source file: every function, struct and macro it defines (document symbols), with kind, signature and #ifdef guards.

Use this INSTEAD of reading a file when you only need its API surface.

Args: file_path: Source file, relative to the project root or absolute.

Inputs: file_path (required), limit, format, offset, +1 more.

reachabilityCall-path reachabilityA
Read-only

Call path / call chain: does FROM transitively call TO, and through which functions?

Use this instead of chaining find_callers rounds when the question is "can this function end up in that one?". BFS over the caller graph returns the SHORTEST chain, each hop with the call site's file:line. JSON results: {path_found, hops, depth, nodes_explored}; hops run from from_symbol to to_symbol. Static analysis cannot follow function pointers (ops structs, callbacks).

Args: from_symbol: The caller end ("can this reach ..."). to_symbol: The callee end ("... this function?"). max_depth: Longest chain to consider, in calls (1-12, default 8).

Inputs: to_symbol (required), from_symbol (required), format, max_depth, +1 more.

update_indexUpdate indexA
Idempotent

Refresh the code index synchronously after editing files — the guaranteed-freshness barrier.

Query tools refresh the index automatically in the background, so results can lag very recent edits by a few seconds. Call this right after editing files when the very next query must see the changes.

Args: full: Rebuild the index from scratch instead of refreshing it incrementally (rarely needed — large branch switch, suspected corruption).

Inputs: full, format, project_root.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.2/5.0

    Scored across 8 tools

    Disambiguation4/5

    Most tools have clearly distinct purposes — definition lookup, reference enumeration, caller/callee analysis, body extraction, file outline, and reachability — and the verbose descriptions reinforce those boundaries. However, find_callers and find_references both surface call sites (one deduplicated by enclosing function, one listing every use), and find_definition overlaps with get_symbol_body enough that an agent could misselect without reading closely.

    Naming Consistency4/5

    Four tools follow a clean find_<plural-noun> pattern (find_callers, find_definition, find_references, find_callees), and list_file_symbols, get_symbol_body, and update_index all use verb_noun as well. The bare noun 'reachability' breaks the pattern, and the mix of 'list'/'get'/'find' verbs is slightly varied, but the overall convention is predictable and readable.

    Tool Count5/5

    Eight tools is well-scoped for a C/C++ code navigation server. Each tool provides a distinct index query — definition, references, callers, callees, call path, symbol body, file outline, and index refresh — with no apparent redundancy, so every tool earns its place.

    Completeness4/5

    The surface covers the core navigation lifecycle thoroughly: definition resolution, reference enumeration, incoming/outgoing call analysis, transitive reachability, body extraction, file-level outlines, and index maintenance. The notable gap is that symbol lookup requires exact names — there is no prefix/fuzzy search for similarly named symbols, even though find_definition's description hints at such a fallback rather than providing a tool for it.

    Maintenance

    ActivityStale
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Provides IDE-like code navigation and search for local repositories, enabling AI assistants to perform symbol search, trigram indexing, and semantic navigation.
      AGPL 3.0
    • A
      license
      Not graded
      quality
      D
      maintenance
      Exposes type-aware code navigation and fast file search to AI agents via language servers, enabling definitions, references, symbols, and file lookup without reading entire codebases.
      2,580 npm
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables LLM agents to efficiently understand and navigate a codebase by providing semantic search over symbols and a reference graph, replacing expensive grep/glob calls with structured tools like definition lookup, caller/callee queries, and change-impact analysis.
      4
      MIT